[ { "run": "alzheimer-apoe4-microglia-ldam-r1", "topic": "alzheimer-apoe4-microglia-ldam", "attempt": 1, "disease": "AD", "census_class": "SUCCESS", "final_status": "done", "n_hypotheses": 6, "counts": { "supported": 0, "partial": 0, "not_supported": 2, "inconclusive": 0, "not_testable": 2, "not_run": 2 }, "evaluated": 4, "stop_reason": null, "question": "Direction d2 from the Alzheimer's-mechanism IDEATE run.\n\nDo APOE4 homozygous and heterozygous carriers in existing human snRNA-seq atlases show a statistically significant shift away from disease-associated microglia (DAM) and toward lipid-droplet- accumulating microglia (LDAM), and does this LDAM/DAM ratio correlate with excess tau pathology burden beyond what is explained by Aβ accumulation alone?\n\nAPOE4 is the strongest genetic risk factor for AD; recent work converges on lipid metabolism and microglial dysfunction as its mechanistic core. TREM2-dependent DAM is the protective microglial state for plaque clearance; LDAM is a dysfunctional, proinflammatory state that fails at phagocytosis. This OmiCairn run re-analyses open human snRNA-seq by APOE genotype, computes DAM vs LDAM microglial composition, and tests whether the LDAM/DAM balance tracks tau burden (Braak / tangle count) independently of amyloid load.\n\nHypotheses (pre-registered, lifted from d2):\n\nH1 (primary — genotype-driven LDAM/DAM marker shift): APOE4/4 carriers show significantly higher LDAM marker expression (LGALS3, CD9, GPNMB, NPC2) and lower DAM2 marker expression (TREM2, LPL, CST7) in microglia compared to APOE3/3 carriers at matched Braak stage, consistent with a lipid-driven microglial dysfunction. Decision rule: per-donor pseudobulk DE within the microglia compartment, APOE4/4 vs APOE3/3 stratified to matched Braak; report log2FC + Benjamini–Hochberg FDR + 95% bootstrap CI per gene. H1 passes if ≥2 of the 4 LDAM markers are up (FDR<0.05) AND ≥2 of the 3 DAM2 markers are down (FDR<0.05).\n\nH2 (secondary — LDAM/DAM ratio vs tau beyond amyloid): The ratio of LDAM to DAM microglial cells correlates positively with neurofibrillary-tangle burden (not just Aβ plaque count), suggesting APOE4-driven microglial dysfunction accelerates tau pathology independently of amyloid load. Decision rule: per-donor LDAM/DAM ratio regressed on a tau index (Braak stage and/or tangle count) with Aβ load as a covariate; report the partial regression coefficient for tau + p-value + 95% CI. H2 passes if the tau coefficient is positive and significant (p<0.05, 95% CI excluding 0) after adjusting for Aβ.\n\nIMPORTANT — covariate dependency: H1 requires APOE genotype metadata; H2 requires tau/Braak (and ideally Aβ) metadata in the same donors. EXECUTE MUST first verify these covariates are present in the fetched object. If APOE genotype is missing, H1 cannot be tested — report that honestly and fall back to a Braak-stratified DAM-vs-LDAM description without the genotype contrast. If tau/Aβ neuropathology is missing, H2 cannot be tested — say so and report the LDAM/DAM composition descriptively. Do NOT impute or fabricate genotype or neuropathology.\n\nA negative or partial result is fully informative. If APOE4 does not shift the LDAM/DAM balance, or the shift does not track tau, that must be the headline rather than re-tuning cluster definitions until H1/H2 pass.", "report": { "kind": "pdf", "url": "/assets/reports/alzheimer-apoe4-microglia-ldam-r1.pdf", "bytes": 330400, "title": "The DAM-GSME Protocol: A Pre-registered Framework for Genotype-Stratified Microglial State Evaluation Reveals LDAM Elevation in APOE4 Carriers Without DAM2 Suppression (SEA-AD MTG, n=84 donors, 39,976 microglia)", "main_conclusion": "Pre-registered H1 (LDAM up / DAM2 down) not supported: composite shift +0.015, 0/7 markers at FDR<0.05, and DAM2 markers trend up rather than down. H2 (LDAM/DAM2 ratio vs Braak, adjusted for Thal) is underpowered and not supported (beta_tau=+0.031, 95% CI -0.007 to 0.069, p=0.143). An exploratory test not adjusted for Braak stage finds a higher LDAM module score in APOE4 carriers (JT=993, p=0.006, Bonferroni p=0.012, r=0.35), which cannot be attributed to genotype.", "thumb": "/assets/img/thumbs/alzheimer-apoe4-microglia-ldam-r1.webp" }, "page": "/campaign/runs/alzheimer-apoe4-microglia-ldam-r1/" }, { "run": "alzheimer-bbb-pericyte-vascular-r1", "topic": "alzheimer-bbb-pericyte-vascular", "attempt": 1, "disease": "AD", "census_class": "PARTIAL", "final_status": "in_progress", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Direction d2 from the second Alzheimer's-mechanism IDEATE run.\n\nDoes pericyte-specific SMAD3 upregulation co-occur with astrocytic VEGFA downregulation in a spatially coherent manner across AD Braak stages, and does this vascular transcriptional programme precede or follow parenchymal amyloid deposition in human brain?\n\nNo prior lab run has examined vascular cells. The \"Gliovascular transcriptional perturbations\" Nature Communications 2024 paper identified pericyte SMAD3 upregulation and astrocytic VEGFA downregulation as the dominant blood-brain-barrier (BBB) disruption signals in AD, with blood SMAD3 levels associating with AD neuroimaging outcomes in a separate human cohort. A companion paper found HIF1A/VEGFA angiogenic failure in AD endothelial cells in the same open dataset (GSE157827, Lau 2020), which is confirmed to contain pericyte, endothelial, and astrocyte populations. This OmiCairn run stratifies donors by Braak stage, applies per-donor pseudobulk DE to vascular cell types, and asks whether the SMAD3/VEGFA axis changes before or after amyloid threshold.\n\nHypotheses (pre-registered, lifted from d2):\n\nH1 (primary — pericyte SMAD3 upregulation by Braak): Pericyte SMAD3 expression is significantly elevated (pseudobulk, FDR < 0.05, fold-change > 1.5) in Braak III-IV vs. Braak 0 donors in GSE157827. Decision rule: per-donor pseudobulk DE within the pericyte compartment, Braak III-IV vs Braak 0; report log2FC + Benjamini-Hochberg FDR + 95% bootstrap CI for SMAD3. H1 passes if SMAD3 is up with fold-change > 1.5 at FDR<0.05 and the 95% CI excludes 0. Report astrocytic VEGFA in parallel (expected down) with the same statistics as a paired vascular-axis readout.\n\nH2 (secondary — vascular programme precedes amyloid threshold): SMAD3 upregulation in pericytes precedes the amyloid plaque density threshold by at least one Braak stage in the GSE157827 donor series, supporting vascular dysfunction as a trigger rather than a consequence. Decision rule: order donors by Braak stage; locate the Braak stage at which pericyte SMAD3 first rises significantly above Braak 0 baseline and the Braak stage at which amyloid plaque density (or its closest available neuropathology proxy) first crosses its threshold; report the Braak-stage offset with a bootstrap 95% CI. H2 passes if the SMAD3-rise stage is at least one Braak stage earlier than the amyloid-threshold stage and the 95% CI of the offset excludes 0.\n\nIMPORTANT — covariate dependency: H1 requires Braak/tau staging in the fetched donors; H2 additionally requires an amyloid plaque-density (or equivalent Aβ) neuropathology measure in the same donors. EXECUTE MUST first verify these are present. If Braak staging is missing, H1 cannot be tested as specified — report that honestly and fall back to an AD-vs-control pericyte SMAD3 contrast without Braak ordering. If amyloid neuropathology is missing, H2 (the precedence question) is untestable — say so and report the Braak-ordered SMAD3 trajectory descriptively without claiming precedence over amyloid. Do NOT impute or fabricate Braak stage or amyloid measures.\n\nA negative or partial result is fully informative. If pericyte SMAD3 is not elevated, or its rise does not precede amyloid, that must be the headline rather than re-tuning cluster definitions until H1/H2 pass.", "report": null, "page": "/campaign/runs/alzheimer-bbb-pericyte-vascular-r1/" }, { "run": "alzheimer-gwas-snatac-celltype-r1", "topic": "alzheimer-gwas-snatac-celltype", "attempt": 1, "disease": "AD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 6, "counts": { "supported": 1, "partial": 0, "not_supported": 3, "inconclusive": 0, "not_testable": 1, "not_run": 1 }, "evaluated": 5, "stop_reason": null, "question": "Direction d3 from the Alzheimer's-mechanism IDEATE run.\n\nCell-type-resolved functional dissection of AD GWAS variants in microglial vs neuronal cis-regulatory elements via snATAC-seq.\n\nWhich of the 75+ AD GWAS loci harbor causal variants within microglia-specific enhancers controlled by SPI1, ELF2, or RUNX1 (vs. excitatory neuron-specific circuits), what are the effector genes and their direction of effect, and do these cell-type-specific regulatory circuits operate differently across AD disease stages in existing multi-omic atlases?\n\nAD GWAS hits are enriched in microglial enhancers (Tsai/Kellis Cell 2023, SPI1/ELF2/RUNX1), but most of the 75+ loci remain functionally uncharacterized at cell-type resolution, and neuronal regulatory circuits may mediate risk through synaptic gene programs independently of microglia. This OmiCairn run re-analyses open human snATAC-seq (the Morabito 2021 multiome ATAC arm) against AD GWAS loci to ask which loci fall in microglia- vs neuron-specific cis-regulatory elements, identify candidate effector genes and the direction of effect, and test whether the microglial regulatory circuits are stage-dependent.\n\nHypotheses (pre-registered, lifted from d3):\n\nH1 (primary — SPI1/microglial enhancer dominance of AD GWAS loci): SPI1 (PU.1) binding motif accessibility in microglia snATAC-seq is the single strongest predictor of causal AD GWAS variant activity, with at least 15 of the top 75 GWAS loci having their strongest chromatin-accessibility QTL (or peak overlap / motif disruption signal) in microglia rather than neurons, astrocytes, or oligodendrocytes. Decision rule: for each AD GWAS lead/credible-set variant, compute cell-type-resolved snATAC peak overlap and SPI1 motif accessibility per cell type; rank cell types per locus; report the count of loci whose strongest microglial signal exceeds the other cell types, with effect size + p-value + 95% CI for the microglia-vs-other enrichment. H1 passes if ≥15 of the top 75 loci are microglia-dominant AND SPI1 motif accessibility is the top predictor (significant at p<0.05) of microglial variant activity.\n\nH2 (secondary — DAM-stage timing of microglial GWAS circuits): Microglial GWAS-variant regulatory circuits are most active at the DAM stage (mid-disease), not at homeostatic or late-disease stages, suggesting the therapeutic window for GWAS-linked microglial targets is the early-to-mid Braak period. Decision rule: stratify microglia by activation state (homeostatic / DAM / late) using marker programs and/or Braak stage; compute GWAS-variant-overlapping peak accessibility per stage; report the stage contrast (DAM vs homeostatic and DAM vs late) as effect size + p-value + 95% CI. H2 passes if peak accessibility at GWAS-overlapping microglial loci is significantly higher at the DAM/mid stage than at both homeostatic and late stages (p<0.05, 95% CI excluding 0).\n\nIMPORTANT — feasibility caveat (snATAC is heavy): snATAC peak-calling and peak-to-variant integration is substantially heavier than the RNA-only AD cases and depends on the snATAC arm being present and parseable in the fetched object, plus a usable AD GWAS variant list. EXECUTE MUST verify, BEFORE asserting H1/H2: (1) the snATAC fragment/ peak matrix is fetchable and parseable; (2) a public AD GWAS lead/credible-set variant list is obtained (ADGC public summary statistics / published supplementary tables — never fabricated accessions or coordinates); (3) variant-to-peak overlap can actually be computed against a stated genome build. If peak calls cannot be computed, the variant overlap cannot be performed, or cell-type labels are absent, EXECUTE MUST report that honestly in analysis summary § Limitations and fall back to whatever partial analysis is supportable (e.g. cell-type marker-gene accessibility, or the RNA arm only) rather than fabricating peak/variant overlaps. Do NOT impute peaks, QTLs, or genotypes.\n\nA negative or partial result is fully informative. If AD GWAS loci are NOT preferentially microglia-/SPI1-resolved, or the circuits are not stage-dependent, that must be the headline rather than re-tuning peak definitions or variant sets until H1/H2 pass.", "report": { "kind": "html", "url": "/campaign/runs/alzheimer-gwas-snatac-celltype-r1/report/", "title": "Microglial enhancers are the dominant chromatin-accessible compartment for AD GWAS lead variants in a prefrontal cortex snATAC-seq atlas, but full locus coverage requires fine-mapped credible sets", "main_conclusion": "Among 33 testable loci (of 75), microglia dominate 13 (39.4%, binomial p=0.048; delta vs other cell types 0.273, p=0.004), but the pre-registered H1 fails under the full 75-locus denominator (13/75, p~0.87). The permutation guard is not significant (p=0.483). H2 (stage-dependent accessibility) is refuted (delta=-0.008, p=0.881).", "thumb": "/assets/img/thumbs/alzheimer-gwas-snatac-celltype-r1.webp" }, "page": "/campaign/runs/alzheimer-gwas-snatac-celltype-r1/" }, { "run": "alzheimer-ligrec-braak-breakdown-r1", "topic": "alzheimer-ligrec-braak-breakdown", "attempt": 1, "disease": "AD", "census_class": "SUCCESS", "final_status": "done", "n_hypotheses": 9, "counts": { "supported": 0, "partial": 0, "not_supported": 1, "inconclusive": 0, "not_testable": 5, "not_run": 3 }, "evaluated": 6, "stop_reason": null, "question": "Direction d5 from the Alzheimer's-mechanism IDEATE run.\n\nWhich ligand-receptor (L-R) communication axes between microglia, astrocytes, and excitatory/inhibitory neurons are most disrupted with advancing Braak stage in human AD snRNA-seq atlases; do the disrupted signals cluster in pathways encoded by AD GWAS risk genes; and can the identity of disrupted signals predict the next region of tau spread (EC→hippocampus→PFC sequence)?\n\nDe Strooper's \"network disease\" framing motivates this: AD emerges from the breakdown of coordinated glial-neuronal responses, not from any single cell-autonomous lesion. Prior work identified dysregulated L-R interactions in AD PFC and integrated GWAS risk into that framework, but did not address the TEMPORAL dynamics across Braak stage. This OmiCairn run quantifies cell-cell communication across Braak stage in open human AD snRNA-seq using open L-R inference frameworks (e.g. CellChat / NicheNet), tests GWAS enrichment of the disrupted axes, and explores whether disrupted signals presage tau spread.\n\nHypotheses (pre-registered, lifted from d5):\n\nH1 (primary — three most-disrupted axes, monotone with Braak): Complement-related signaling (C1Q-C3-CR3 axis: microglia→neuron), SPP1 (microglia→astrocyte/oligodendrocyte), and CXCL10-CXCR3 (glia→T-cell recruitment) are the three most disrupted communication axes in AD, with disruption magnitude increasing monotonically with Braak stage. Decision rule: per-donor pseudobulk L-R interaction scores (CellChat communication probability or equivalent) regressed on Braak stage (ordinal); report slope + p-value + 95% CI per axis. H1 passes if the three named axes rank among the top disrupted interactions AND each shows a positive monotone slope with Braak (p<0.05, 95% CI excluding 0). Report the full ranked disruption table regardless — if a different axis dominates, that is the finding.\n\nH2 (secondary — GWAS enrichment of disrupted axes): Disrupted L-R interactions at Braak stage III-IV are significantly enriched in pathways encoded by genes in the top 20 AD GWAS loci (specifically BIN1, CLU, CR1, MS4A, PICALM, TREM2), indicating that GWAS-identified risk converges on inter-cellular communication failure rather than purely cell-autonomous dysfunction. Decision rule: hypergeometric / permutation enrichment test of the Braak III-IV disrupted-interaction gene set against the top-20 AD GWAS locus gene set; report odds ratio / enrichment statistic + p-value + 95% CI. H2 passes if enrichment is significant (p<0.05) with a 95% CI excluding the null.\n\nA negative or partial result is fully informative. If complement / SPP1 / CXCL10 are NOT the top disrupted axes, report the actual ranking as the headline. If GWAS enrichment is absent, report that the disrupted communication network does not preferentially involve GWAS loci. Do not re-tune the L-R inference until H1/H2 pass.\n\nThe tau-spread prediction (EC→hippocampus→PFC) is exploratory and multi-region: with a single-region primary atlas it can only be framed as a hypothesis-generating direction, not tested. Report it as such unless a multi-region open atlas (SEA-AD) is successfully fetched and harmonised.", "report": { "kind": "pdf", "url": "/assets/reports/alzheimer-ligrec-braak-breakdown-r1.pdf", "bytes": 380975, "title": "LIGREC-BRAAK-AUDIT: A standardized ligand–receptor disruption protocol reveals ECM–integrin dominance in Alzheimer's prefrontal cortex with null complement and GWAS signals", "main_conclusion": "The global L-R disruption metric is indistinguishable from the donor-label permutation null (top-10 mean |log2FC| 1.915 vs null 1.913, p=0.494; no pair at FDR<0.05). This is an underpowered null at n=21. H1 fails on all pre-registered axes (complement rank 257, SPP1 reduced, CXCL10-CXCR3 not testable) and H2 GWAS enrichment is OR=0 (p=1.0, structural absence from CellChatDB). ECM-integrin pairs rank highest only numerically (delta=0.662 vs named axes, p=0.001).", "thumb": "/assets/img/thumbs/alzheimer-ligrec-braak-breakdown-r1.webp" }, "page": "/campaign/runs/alzheimer-ligrec-braak-breakdown-r1/" }, { "run": "alzheimer-rorb-ec-vulnerability-r1", "topic": "alzheimer-rorb-ec-vulnerability", "attempt": 1, "disease": "AD", "census_class": "SUCCESS", "final_status": "done", "n_hypotheses": 8, "counts": { "supported": 0, "partial": 0, "not_supported": 4, "inconclusive": 2, "not_testable": 0, "not_run": 2 }, "evaluated": 6, "stop_reason": null, "question": "Direction d1 from the Alzheimer's-mechanism IDEATE run.\n\nWhat cell-intrinsic transcriptional programs — specifically excitability regulation, calcium handling, proteostasis capacity, and mitochondrial gene expression — distinguish selectively vulnerable RORB+ entorhinal cortex (EC) layer II excitatory neurons from resilient neurons in the same region, and can these molecular features predict inter-individual variation in tau accumulation rate across Braak stages using existing OPEN snRNA-seq atlases?\n\nEC layer II neurons are the first to accumulate phospho-tau and die, defining Braak stage I–II. Leng et al. (2021, Nat Neurosci) showed RORB expression marks these vulnerable neurons with ~50% abundance loss at early stages. WHY RORB+ neurons are vulnerable — high intrinsic excitability, low mitochondrial capacity, limited proteostasis reserve, or a combination — remains unresolved. This OmiCairn run re-analyses open EC snRNA-seq atlases to test candidate cell-intrinsic vulnerability mechanisms by contrasting RORB+ vs. RORB- excitatory neurons within region and across Braak stage.\n\nHypotheses (pre-registered, lifted from d1):\n\nH1 (primary — calcium-handling / excitability signature): RORB+ EC layer II excitatory neurons express significantly higher levels of voltage-gated calcium-channel subunits (CACNA1C, CACNA1D) and lower levels of calcium-buffering genes (CALB1, PVALB) relative to resilient RORB- EC excitatory neurons, consistent with a calcium-/excitability- driven basis for their selective tau vulnerability. Decision rule: per-donor pseudobulk DE within EC excitatory neurons, RORB+ vs RORB-; report log2FC + Benjamini–Hochberg FDR + 95% bootstrap CI for each gene. H1 passes if CACNA1C and/or CACNA1D are up (log2FC > 0, FDR < 0.05) AND CALB1 and/or PVALB are down (log2FC < 0, FDR < 0.05) in the primary atlas, with the directional sign reproduced in the Braak-staged secondary atlas.\n\nH2 (secondary — integrated stress response / SASP precursor trajectory): Across Braak stages I–IV in EC snRNA-seq data, RORB+ neurons show a progressive shift toward an integrated stress response (ATF4-high; phospho-eIF2α surrogate genes — e.g. ATF4, DDIT3/CHOP, PPP1R15A, ATF3, TRIB3 — upregulated) and a senescence-associated secretory (SASP) precursor state, detectable before frank neuronal loss. Decision rule: ISR/SASP module score per RORB+ cell regressed on Braak stage (ordinal); report slope + p-value + 95% CI. H2 passes if the ISR module score increases monotonically with Braak stage in RORB+ neurons (positive slope, p < 0.05) and the increase precedes the inflection point of RORB+ abundance loss. Report the RORB+ abundance-vs-Braak curve alongside.\n\nA negative or partial result is fully informative. If the calcium / excitability contrast does not separate RORB+ from RORB- neurons, or if the ISR shift is absent, that finding must be reported as the headline rather than re-tuning clustering or gene panels until H1/H2 pass. Report which decomposition (excitability vs proteostasis vs mitochondrial) carries the signal.", "report": { "kind": "pdf", "url": "/assets/reports/alzheimer-rorb-ec-vulnerability-r1.pdf", "bytes": 331257, "title": "RORB+ entorhinal cortex layer II neurons carry a broad excitability signature of uncertain donor-level precision, independent of CACNA1C/D", "main_conclusion": "Powered negatives: CACNA1C/D are not upregulated in RORB+ vs RORB- EC excitatory neurons (log2FC -0.026 / -0.065, BH-FDR 1.00; H1 not supported), and the ISR score does not rise with Braak stage (slope -0.006, p=0.213; H2 not supported). A broad excitability module points upward (d=+0.269), but its donor-level CI [-0.362, +0.900] spans zero.", "thumb": "/assets/img/thumbs/alzheimer-rorb-ec-vulnerability-r1.webp" }, "page": "/campaign/runs/alzheimer-rorb-ec-vulnerability-r1/" }, { "run": "ideate-alzheimer-mechanism-r1", "topic": "ideate-alzheimer-mechanism", "attempt": 1, "disease": "AD", "census_class": "EARLY", "final_status": "halted", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "IDEATE_ONLY", "question": "Research purpose: understand the MECHANISM of Alzheimer's disease (AD) — the molecular, cellular, circuit, and systems-level processes that drive selective neuronal vulnerability and disease progression (amyloid-β, tau/NFTs, APOE & genetics/GWAS, microglia & neuroinflammation, synaptic loss, vascular/BBB dysfunction, lysosomal-autophagy, astrocytes) — and propose PROMISING FUTURE RESEARCH DIRECTIONS that are concrete enough for an autonomous computational pipeline to pursue on PUBLIC data.\n\nTargets:\n\n(1) Survey AD-mechanism evidence broadly: papers (single-cell/spatial atlases, genetics/GWAS, proteomics, amyloid-β biology, tau/NFT pathology, microglia/neuroinflammation, synaptic loss, vascular/BBB, lysosomal-autophagy, astrocytes, selective neuronal vulnerability e.g. entorhinal cortex / EC), plus stated positions from leading AD researchers and relevant philosophy-of-medicine views on causation/biomarkers (e.g. the amyloid-cascade debate).\n\n(2) Propose directions that are NOVEL relative to the campaign's existing topics — do not re-propose work already covered; find the gaps and the cross-cutting syntheses between them.\n\n(3) For each proposed direction, identify CANDIDATE PUBLIC DATASETS (named atlases / accessions / repositories: GEO, SRA, PRIDE, Synapse, cellxgene, EBI, Allen, ADNI-derived public tables, etc.) that the direction would use. Put dataset pointers in the direction's rationale.\n\nIMPORTANT — downloadability: prefer OPENLY-DOWNLOADABLE datasets (GEO supplementary, Synapse-open, cellxgene, EBI) so that the downstream EXECUTE stage can actually fetch the data. For any dataset that is access-controlled (dbGaP, controlled ROSMAP, controlled SEA-AD), EXPLICITLY FLAG it as access-controlled, since fetching it requires an application. As examples it may consider, well-known OPEN human AD single-cell/transcriptomic datasets (treat as candidates and VERIFY accessions — do not assume): Mathys 2019 (prefrontal cortex snRNA-seq), Grubman 2019 (candidate GSE138852, entorhinal cortex), Lau 2020 (candidate GSE157827, prefrontal cortex), Leng 2021 (entorhinal + superior frontal gyrus), Morabito 2021 (candidate GSE174367, snRNA + snATAC), Gerrits 2021, Sayed 2021.\n\nThis is a survey + proposal task, not a data-analysis task.", "report": null, "page": "/campaign/runs/ideate-alzheimer-mechanism-r1/" }, { "run": "ideate-alzheimer-mechanism-r2", "topic": "ideate-alzheimer-mechanism", "attempt": 2, "disease": "AD", "census_class": "EARLY", "final_status": "halted", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "IDEATE_ONLY", "question": "Research purpose: understand the MECHANISM of Alzheimer's disease (AD) — the molecular, cellular, circuit, and systems-level processes that drive selective neuronal vulnerability and disease progression (amyloid-β, tau/NFTs, APOE & genetics/GWAS, microglia & neuroinflammation, synaptic loss, vascular/BBB dysfunction, lysosomal-autophagy, astrocytes) — and propose PROMISING FUTURE RESEARCH DIRECTIONS that are concrete enough for an autonomous computational pipeline to pursue on PUBLIC data.\n\nTargets:\n\n(1) Survey AD-mechanism evidence broadly: papers (single-cell/spatial atlases, genetics/GWAS, proteomics, amyloid-β biology, tau/NFT pathology, microglia/neuroinflammation, synaptic loss, vascular/BBB, lysosomal-autophagy, astrocytes, selective neuronal vulnerability e.g. entorhinal cortex / EC), plus stated positions from leading AD researchers and relevant philosophy-of-medicine views on causation/biomarkers (e.g. the amyloid-cascade debate).\n\n(2) Propose directions that are NOVEL relative to the campaign's existing topics — do not re-propose work already covered; find the gaps and the cross-cutting syntheses between them.\n\n(3) For each proposed direction, identify CANDIDATE PUBLIC DATASETS (named atlases / accessions / repositories: GEO, SRA, PRIDE, Synapse, cellxgene, EBI, Allen, ADNI-derived public tables, etc.) that the direction would use. Put dataset pointers in the direction's rationale.\n\nIMPORTANT — downloadability: prefer OPENLY-DOWNLOADABLE datasets (GEO supplementary, Synapse-open, cellxgene, EBI) so that the downstream EXECUTE stage can actually fetch the data. For any dataset that is access-controlled (dbGaP, controlled ROSMAP, controlled SEA-AD), EXPLICITLY FLAG it as access-controlled, since fetching it requires an application. As examples it may consider, well-known OPEN human AD single-cell/transcriptomic datasets (treat as candidates and VERIFY accessions — do not assume): Mathys 2019 (prefrontal cortex snRNA-seq), Grubman 2019 (candidate GSE138852, entorhinal cortex), Lau 2020 (candidate GSE157827, prefrontal cortex), Leng 2021 (entorhinal + superior frontal gyrus), Morabito 2021 (candidate GSE174367, snRNA + snATAC), Gerrits 2021, Sayed 2021.\n\nThis is a survey + proposal task, not a data-analysis task.", "report": null, "page": "/campaign/runs/ideate-alzheimer-mechanism-r2/" }, { "run": "ideate-parkinson-mechanism-r1", "topic": "ideate-parkinson-mechanism", "attempt": 1, "disease": "PD", "census_class": "EARLY", "final_status": "halted", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "IDEATE_ONLY", "question": "Research purpose: understand the MECHANISM of Parkinson's disease (PD) — the molecular, cellular, circuit, and systems-level processes that drive dopaminergic neuron vulnerability and disease progression — and propose PROMISING FUTURE RESEARCH DIRECTIONS that are concrete enough for an autonomous computational pipeline to pursue on PUBLIC data.\n\nTargets:\n\n(1) Survey PD-mechanism evidence broadly: papers (single-cell/spatial atlases, genetics/GWAS, proteomics, alpha-synuclein biology, mitochondrial/lysosomal pathways, neuroinflammation, gut-brain axis), plus stated positions from leading PD researchers and relevant philosophy-of-medicine views on causation/biomarkers.\n\n(2) Propose directions that are NOVEL relative to the campaign's existing topics — do not re-propose work already covered; find the gaps and the cross-cutting syntheses between them.\n\n(3) For each proposed direction, identify CANDIDATE PUBLIC DATASETS (named atlases / accessions / repositories: GEO, SRA, PRIDE, Allen, PPMI-derived public tables, etc.) that the direction would use. Put dataset pointers in the direction's rationale.\n\nThis is a survey + proposal task, not a data-analysis task.", "report": null, "page": "/campaign/runs/ideate-parkinson-mechanism-r1/" }, { "run": "ideate-parkinson-mechanism-r2", "topic": "ideate-parkinson-mechanism", "attempt": 2, "disease": "PD", "census_class": "SUCCESS", "final_status": "done", "n_hypotheses": 6, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 1, "not_testable": 5, "not_run": 0 }, "evaluated": 6, "stop_reason": null, "question": "Research purpose: understand the MECHANISM of Parkinson's disease (PD) — the molecular, cellular, circuit, and systems-level processes that drive dopaminergic neuron vulnerability and disease progression — and propose PROMISING FUTURE RESEARCH DIRECTIONS that are concrete enough for an autonomous computational pipeline to pursue on PUBLIC data.\n\nTargets:\n\n(1) Survey PD-mechanism evidence broadly: papers (single-cell/spatial atlases, genetics/GWAS, proteomics, alpha-synuclein biology, mitochondrial/lysosomal pathways, neuroinflammation, gut-brain axis), plus stated positions from leading PD researchers and relevant philosophy-of-medicine views on causation/biomarkers.\n\n(2) Propose directions that are NOVEL relative to the campaign's existing topics — do not re-propose work already covered; find the gaps and the cross-cutting syntheses between them.\n\n(3) For each proposed direction, identify CANDIDATE PUBLIC DATASETS (named atlases / accessions / repositories: GEO, SRA, PRIDE, Allen, PPMI-derived public tables, etc.) that the direction would use. Put dataset pointers in the direction's rationale.\n\nThis is a survey + proposal task, not a data-analysis task.", "report": { "kind": "pdf", "url": "/assets/reports/ideate-parkinson-mechanism-r2.pdf", "bytes": 268145, "title": "LYDRA: A Pre-Registered Protocol for Testing Lysosomal Convergence Hypotheses in Parkinson's Disease iPSC Transcriptomics — Pilot Results and Data-Access Audit", "main_conclusion": "H1 is blocked (no dbGaP access to the Kamath 2022 SN atlas) and H3 is blocked (no matched RNA/proteomics data). H2 used a post-hoc V-ATPase surrogate: mean log2FC +0.485 (95% CI -0.454 to +1.471; permutation p=0.333; N=2 donors per group). It points the right way but is underpowered, so H2 is NOT SUPPORTED; the report is mainly a protocol and data-access audit.", "thumb": "/assets/img/thumbs/ideate-parkinson-mechanism-r2.webp" }, "page": "/campaign/runs/ideate-parkinson-mechanism-r2/" }, { "run": "ideate-parkinson-mechanism-r3", "topic": "ideate-parkinson-mechanism", "attempt": 3, "disease": "PD", "census_class": "EARLY", "final_status": "halted", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "IDEATE_ONLY", "question": "Research purpose: understand the MECHANISM of Parkinson's disease (PD) — the molecular, cellular, circuit, and systems-level processes that drive dopaminergic neuron vulnerability and disease progression — and propose PROMISING FUTURE RESEARCH DIRECTIONS that are concrete enough for an autonomous computational pipeline to pursue on PUBLIC data.\n\nTargets:\n\n(1) Survey PD-mechanism evidence broadly: papers (single-cell/spatial atlases, genetics/GWAS, proteomics, alpha-synuclein biology, mitochondrial/lysosomal pathways, neuroinflammation, gut-brain axis), plus stated positions from leading PD researchers and relevant philosophy-of-medicine views on causation/biomarkers.\n\n(2) Propose directions that are NOVEL relative to the campaign's existing topics — do not re-propose work already covered; find the gaps and the cross-cutting syntheses between them.\n\n(3) For each proposed direction, identify CANDIDATE PUBLIC DATASETS (named atlases / accessions / repositories: GEO, SRA, PRIDE, Allen, PPMI-derived public tables, etc.) that the direction would use. Put dataset pointers in the direction's rationale.\n\nThis is a survey + proposal task, not a data-analysis task.", "report": null, "page": "/campaign/runs/ideate-parkinson-mechanism-r3/" }, { "run": "neurodeg-microglia-convergence-r1", "topic": "neurodeg-microglia-convergence", "attempt": 1, "disease": "CROSS", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 6, "counts": { "supported": 0, "partial": 3, "not_supported": 2, "inconclusive": 1, "not_testable": 0, "not_run": 0 }, "evaluated": 6, "stop_reason": null, "question": "Extension of two prior OmiCairn lab-KB findings. (1) A Jonckheere–Terpstra test on the SEA-AD Microglia-PVM MTG atlas (CellxGene collection 1ca90a2d) found the LDAM module score (LGALS3, CD9, GPNMB, NPC2) significantly elevated in APOE4 carriers vs non-carriers (JT=993.0, p=0.006), while the DAM2 score (TREM2, LPL, CST7) showed no significant carrier shift; a follow-up tau regression was directionally positive but null (β_tau=+0.031, p=0.143, 95% CI [−0.007, +0.069]). (2) In AD prefrontal cortex (GSE157827, 21 donors) the most-disrupted ligand–receptor pairs were ECM–integrin interactions (LAMA2→ITGAV/ITGA6, COL4A2→ITGB8; microglia/endothelial→astrocyte; log2FC ≈ −2.0). Both findings were observed within Alzheimer's disease alone. The open AD/DLB/PDD prefrontal-cortex atlas GSE303823 (Brase/Harari-style cohort; processed object on Zenodo 11107597) independently reports a shared, myelin-enriched microglial state with increased MSR1-mediated phagocytosis across all three dementia subtypes.\n\nThis run asks whether the APOE4-linked lipid-droplet microglial program (LDAM) is a SHARED cross-disease feature of dementia or an AD-specific one, and whether the ECM–integrin disruption is, by contrast, AD-specific. Primary discovery cohort: GSE303823 (snRNA-seq PFC; NCI control n=4, AD n=4, DLB n=6, PDD n=6; ~20 donors, the processed `rna_list.rds` on Zenodo 11107597). Orthogonal AD-arm replication: SEA-AD on CellxGene (open; MTG + DLPFC; the Microglia-PVM object under collection 1ca90a2d, plus the full atlas). Held-out blind external validation: SEA-AD DLPFC (A9) — never used to define the microglial states or fit any model.\n\nHypotheses (pre-registered):\n\nH1 (primary — shared vs AD-specific LDAM elevation across 3 diseases): The LDAM module score (LGALS3, CD9, GPNMB, NPC2) within the microglial compartment is elevated in EACH of AD, DLB and PDD relative to NCI controls in GSE303823, i.e. it is a SHARED dementia feature rather than AD-restricted. Decision rule: per-donor pseudobulk LDAM module score in microglia; for each disease vs NCI report Hedges' g effect size + p-value + 95% bootstrap CI, with Benjamini–Hochberg FDR across the 3 contrasts. H1 (shared) passes if all 3 diseases show a positive elevation with FDR<0.05 and 95% CI excluding 0. It is DECISIVELY classed AD-specific (the competing model) if only AD is significant and DLB/PDD CIs include 0. A partial pattern (e.g. AD+DLB but not PDD) is reported verbatim as partial sharing — NOT rounded up to \"shared\".\n\nH2 (secondary — orthogonal method corroboration): The shared/AD-specific LDAM pattern in H1 is reproduced by an independent quantification: (a) compositional shift of an unsupervised LDAM-like microglial subcluster (Leiden, scored post-hoc by the LDAM markers) tested with a Dirichlet/scCODA-style per-donor proportion model, and (b) SCENIC/regulon or AUCell module activity for the same gene set. H2 corroborates H1 if the sign and the shared-vs-specific call agree across at least 2 of the 3 diseases under both orthogonal methods.\n\nH3 (secondary — held-out blind external validation, AD arm only): A signature locked from GSE303823 (the per-gene LDAM weights defining the microglial program) is scored, WITHOUT refitting, on the held-out SEA-AD DLPFC (A9) atlas to test whether it separates high- vs low-pathology AD donors. Report AUC + 1000× bootstrap 95% CI. The lock file (gene list + weights + cluster definition) is frozen before touching SEA-AD DLPFC. A null AUC (CI spanning 0.5) is an informative, reportable outcome.\n\nH4 (secondary — APOE4-specificity of the lipid-droplet sub-program): IF APOE genotype / APOE4-carrier status is present in GSE303823 and/or SEA-AD donors, the lipid-droplet sub-program (the LGALS3/GPNMB-high axis) is enriched specifically in APOE4 carriers, replicating the prior JT=993.0, p=0.006 SEA-AD finding in an independent cohort. Decision rule: carrier vs non-carrier Jonckheere–Terpstra (or Mann–Whitney) on the per-donor sub-program score, with effect size + p + 95% CI. COVARIATE DEPENDENCY: if APOE status is absent, H4 is UNTESTABLE — report that honestly and fall back to a genotype-agnostic description. Do NOT impute genotype.\n\nH5 (secondary — NEGATIVE CONTROL: ECM–integrin disruption is AD-specific, not shared): The ECM–integrin ligand–receptor disruption observed in AD PFC (LAMA2→ITGAV/ITGA6, COL4A2→ITGB8; microglia/endothelial→ astrocyte) is AD-specific and is NOT shared by DLB or PDD. Decision rule: per-donor pseudobulk expression of the ECM ligands (LAMA2, COL4A2, and the integrin receptors) and/or a CellChat-style L-R disruption score, each disease vs NCI; report effect size + p + 95% CI. This is a pre-registered DISSOCIATION test: H5 passes (supporting the convergence story) if AD shows the disruption while DLB/PDD do NOT. If the ECM disruption is ALSO shared across diseases, that is an equally informative negative-control failure and must be reported as the headline finding, not buried.\n\nH6 (secondary — honest-failure / sensitivity decomposition): Decompose the H1 LDAM effect into nuisance vs real-biology layers: (i) microglial-proportion differences across disease (does the score shift just track more microglia?), (ii) post-mortem-interval / age / sex confounds (covariate- adjusted re-fit), (iii) Leiden-resolution sweep and marker-set perturbation (swap to the expanded LDAM set adding APOE/CLU, per the prior sensitivity finding). State explicitly that resolution-instability or a proportion-driven artifact is itself the headline if it dominates.\n\nA negative or partial result is fully informative. If the LDAM elevation turns out AD-specific (not shared), or the ECM negative control fails (disruption is shared after all), or H3 validation is null, THAT is the headline result. The analysis must NOT be retuned (re-clustering, marker-set cherry-picking) until H1 reads \"shared\".", "report": { "kind": "html", "url": "/campaign/runs/neurodeg-microglia-convergence-r1/report/", "title": "Lewy body dementias, not Alzheimer's disease, preferentially drive the lipid-droplet microglial program: a pre-registered cross-disease snRNA-seq analysis", "main_conclusion": "LDAM is FDR-significant in DLB (g=1.55) and PDD (g=2.64) but not AD (g=1.20, FDR=0.343), giving a 'partial sharing' call. But DAM2 co-elevates the same way, held-out SEA-AD validation is null (AUC=0.60, CI spans 0.5), APOE4-LDAM does not replicate, and H5 ECM is inconclusive, so the result points to generic Lewy-body microglial activation rather than a lipid-droplet-specific program.", "thumb": "/assets/img/thumbs/neurodeg-microglia-convergence-r1.webp" }, "page": "/campaign/runs/neurodeg-microglia-convergence-r1/" }, { "run": "parkinson-asyn-propagation-r1", "topic": "parkinson-asyn-propagation", "attempt": 1, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 2, "partial": 0, "not_supported": 1, "inconclusive": 2, "not_testable": 0, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Does a Susceptible-Infected-Removed (SIR) compartmental model of α-synuclein spread between brain regions, parameterised by Fullard 2024 multi-region snRNA-seq cell-type fractions and SNCA expression, plus Allen Human Brain Connectivity Atlas projection edges, reproduce the canonical Braak staging order (DMNX → SNpc → cingulate → M1 → V1) at qualitatively correct relative timing?\n\nBraak staging (Braak 2003 Neurobiol Aging) is the most-cited spatial-temporal claim in PD neuropathology, but its mechanistic explanation — that α-syn spreads cell-to-cell along a connectome gradient, modulated by selective vulnerability — has been tested in mouse atlases (Henderson 2019, Zheng 2019) but not at human cell-type granularity. Fullard 2024 provides the missing per-region cell-fraction and SNCA-expression priors.\n\nHypotheses (pre-registered):\n\nH1 (primary — Braak rank-order recovery): An SIR + connectivity model with SNCA-expression-weighted per-region susceptibility recovers the rank-order of predicted regional peak-prevalence times consistent with Braak rank, with Spearman ρ ≥ 0.7 across the 5 modelled regions (DMNX, SNpc, cingulate, M1, V1). Bootstrap 95% CI of ρ excludes 0.\n\nH2 (secondary — SNCA-weighting necessity): Ablating the SNCA-expression weighting (uniform susceptibility across regions) degrades rank-order recovery by ≥ 0.3 in Spearman ρ. (Test that cell-specific susceptibility is necessary, not sufficient.)\n\nH3 (secondary — cingulate cell-type prediction): The model predicts which cingulate cell types reach infection threshold first; we pre-register \"excitatory layer-5/6 neurons\" as the top prediction. The top-3 model-predicted cell types overlap ≥ 2 with the top-3 Lewy- burdened cell types reported by Goralski 2024 LAMDA (Zenodo 10.5281/zenodo.10729767).\n\nH4 (secondary — vagal-route ablation): Removing the DMNX → SN projection edge (parameter ablation) shifts predicted peak-time rank-order by > 1 rank position. A falsifiable mechanistic claim about the vagal spread route.\n\nH5 (secondary — connectome-only vs. cell-autonomous): Compare three models — (A) connectome-only SIR, no cell-type weighting; (B) cell-type-weighted SIR (this proposal's primary); (C) cell-autonomous-only (each region's I trajectory independent of others). Report which best fits Braak rank-order. A finding that (A) or (C) wins is an informative inversion of the spread-only hypothesis.\n\nNull payoff. If no SIR variant recovers Braak rank-order, the connectome-only hypothesis is incomplete and pure cell-autonomous selective vulnerability (independent of spread) dominates. Informative as a head-to-head test of two competing field-level hypotheses.", "report": null, "page": "/campaign/runs/parkinson-asyn-propagation-r1/" }, { "run": "parkinson-asyn-propagation-r2", "topic": "parkinson-asyn-propagation", "attempt": 2, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 0, "not_supported": 4, "inconclusive": 1, "not_testable": 0, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Does a Susceptible-Infected-Removed (SIR) compartmental model of α-synuclein spread between brain regions, parameterised by Fullard 2024 multi-region snRNA-seq cell-type fractions and SNCA expression, plus Allen Human Brain Connectivity Atlas projection edges, reproduce the canonical Braak staging order (DMNX → SNpc → cingulate → M1 → V1) at qualitatively correct relative timing?\n\nBraak staging (Braak 2003 Neurobiol Aging) is the most-cited spatial-temporal claim in PD neuropathology, but its mechanistic explanation — that α-syn spreads cell-to-cell along a connectome gradient, modulated by selective vulnerability — has been tested in mouse atlases (Henderson 2019, Zheng 2019) but not at human cell-type granularity. Fullard 2024 provides the missing per-region cell-fraction and SNCA-expression priors.\n\nHypotheses (pre-registered):\n\nH1 (primary — Braak rank-order recovery): An SIR + connectivity model with SNCA-expression-weighted per-region susceptibility recovers the rank-order of predicted regional peak-prevalence times consistent with Braak rank, with Spearman ρ ≥ 0.7 across the 5 modelled regions (DMNX, SNpc, cingulate, M1, V1). Bootstrap 95% CI of ρ excludes 0.\n\nH2 (secondary — SNCA-weighting necessity): Ablating the SNCA-expression weighting (uniform susceptibility across regions) degrades rank-order recovery by ≥ 0.3 in Spearman ρ. (Test that cell-specific susceptibility is necessary, not sufficient.)\n\nH3 (secondary — cingulate cell-type prediction): The model predicts which cingulate cell types reach infection threshold first; we pre-register \"excitatory layer-5/6 neurons\" as the top prediction. The top-3 model-predicted cell types overlap ≥ 2 with the top-3 Lewy- burdened cell types reported by Goralski 2024 LAMDA (Zenodo 10.5281/zenodo.10729767).\n\nH4 (secondary — vagal-route ablation): Removing the DMNX → SN projection edge (parameter ablation) shifts predicted peak-time rank-order by > 1 rank position. A falsifiable mechanistic claim about the vagal spread route.\n\nH5 (secondary — connectome-only vs. cell-autonomous): Compare three models — (A) connectome-only SIR, no cell-type weighting; (B) cell-type-weighted SIR (this proposal's primary); (C) cell-autonomous-only (each region's I trajectory independent of others). Report which best fits Braak rank-order. A finding that (A) or (C) wins is an informative inversion of the spread-only hypothesis.\n\nNull payoff. If no SIR variant recovers Braak rank-order, the connectome-only hypothesis is incomplete and pure cell-autonomous selective vulnerability (independent of spread) dominates. Informative as a head-to-head test of two competing field-level hypotheses.", "report": { "kind": "pdf", "url": "/assets/reports/parkinson-asyn-propagation-r2.pdf", "bytes": 355063, "title": "Connectome topology (rho = 1.00) outperforms SNCA expression for Braak alpha-synuclein staging: a pre-registered human atlas benchmark", "main_conclusion": "The pre-registered SNCA-expression-weighted SIR model failed H1 (rho = 0.711, 95% CI -0.304 to 1.000, permutation p = 0.094, n = 5 regions), and H2, H3 and H4 were also not supported. Connectome-only models did better (rho = 0.918 at shared parameters; rho = 1.000 when optimised post hoc), and Dijkstra graph distance alone gave rho = 0.975 (p = 0.015). The report reads this as the Allen connectivity topology encoding the Braak gradient, a confound for atlas-based propagation analyses.", "thumb": "/assets/img/thumbs/parkinson-asyn-propagation-r2.webp" }, "page": "/campaign/runs/parkinson-asyn-propagation-r2/" }, { "run": "parkinson-cellchat-immune-axis-r1", "topic": "parkinson-cellchat-immune-axis", "attempt": 1, "disease": "PD", "census_class": "PARTIAL", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 5 }, "evaluated": 0, "stop_reason": null, "question": "Do LIANA-py (CellChat-equivalent consensus) inferred ligand- receptor (L-R) communications between Hong 2025 peripheral-blood CD8+ T cells (GEO GSE223138) and Fullard brain microglia mirror — at the L-R pair level — the brain-internal CD8+T → microglia communications inferred by Ma 2025 (PMC12322057), thus identifying a \"blood-as-foreshadow-of-brain\" set of peripheral biomarker ligands?\n\nHong 2025 (npj Parkinson's 11) reported 28 immune-cell-specific PD risk genes from OneK1K + IPDGC + GSE223138 multi-omic integration. Ma 2025 (Nat Commun 16) characterised brain-side CD8+ T-cell × microglia communications via SN scRNA + scTCR-seq. The two cell-type-communication signatures have not been compared at the L-R pair level. If peripheral L-Rs echo brain L-Rs, the peripheral signal could be used as a foreshadow blood biomarker; if they diverge, peripheral-blood-as-PD-biomarker biology is decoupled from the brain-immune axis. Either outcome is informative with strong biomarker implications.\n\nHypotheses (pre-registered):\n\nH1 (primary — L-R pair concordance): ≥ 5 ligand-receptor pairs are inferred as PD-up-regulated in BOTH the peripheral CD8+T → blood-monocyte direction (Hong 2025) AND the brain CD8+T → microglia direction (Ma 2025), with concordant sign of dysregulation. Permutation null p < 0.01 (10,000 random L-R subsets).\n\nH2 (secondary — Hong-MR-gene enrichment): The shared L-R pairs are enriched (q < 0.05 hypergeometric) for ligands or receptors whose genes appear on the Hong 2025 MR-implicated immune-gene list (FDFT1, CTSB, ARSA, HLA-DQA1, and 24 others reported in Hong 2025 Table 2).\n\nH3 (secondary — peripheral-foreshadow correlation): A \"peripheral CD8+T signature score\" computed from Hong 2025 PBMC data using the 28 MR-implicated immune genes rank-correlates (Spearman ρ > 0.3, nominal p < 0.05) with the v1 `parkinson-tcell-foothold` per-donor brain T-cell-module score at the disease-stage level (early vs. late PD).\n\nH4 (secondary — direction-specific dropout): Inferring CD8+T → monocyte (PBMC) and microglia → CD8+T (brain) — the reverse-direction pairs — should NOT show the same H1 enrichment (control: PD signals are in the CD8+T → myeloid direction, not the reverse).\n\nH5 (secondary — method-consensus robustness): Repeat the L-R inference with three methods (CellPhoneDB, NATMI, scTalk) and confirm the H1 shared-pair count is within ±2 of the LIANA consensus result. (LIANA-py provides this consensus natively.)\n\nNull payoff. If peripheral and brain L-R signatures diverge (H1: 0-3 shared pairs), peripheral-blood-as-PD-biomarker biology is decoupled from the brain-immune axis — informative, with strong implications for blood-test development.", "report": null, "page": "/campaign/runs/parkinson-cellchat-immune-axis-r1/" }, { "run": "parkinson-cellchat-immune-axis-r2", "topic": "parkinson-cellchat-immune-axis", "attempt": 2, "disease": "PD", "census_class": "PARTIAL", "final_status": "in_progress", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Do LIANA-py (CellChat-equivalent consensus) inferred ligand- receptor (L-R) communications between Hong 2025 peripheral-blood CD8+ T cells (GEO GSE223138) and Fullard brain microglia mirror — at the L-R pair level — the brain-internal CD8+T → microglia communications inferred by Ma 2025 (PMC12322057), thus identifying a \"blood-as-foreshadow-of-brain\" set of peripheral biomarker ligands?\n\nHong 2025 (npj Parkinson's 11) reported 28 immune-cell-specific PD risk genes from OneK1K + IPDGC + GSE223138 multi-omic integration. Ma 2025 (Nat Commun 16) characterised brain-side CD8+ T-cell × microglia communications via SN scRNA + scTCR-seq. The two cell-type-communication signatures have not been compared at the L-R pair level. If peripheral L-Rs echo brain L-Rs, the peripheral signal could be used as a foreshadow blood biomarker; if they diverge, peripheral-blood-as-PD-biomarker biology is decoupled from the brain-immune axis. Either outcome is informative with strong biomarker implications.\n\nHypotheses (pre-registered):\n\nH1 (primary — L-R pair concordance): ≥ 5 ligand-receptor pairs are inferred as PD-up-regulated in BOTH the peripheral CD8+T → blood-monocyte direction (Hong 2025) AND the brain CD8+T → microglia direction (Ma 2025), with concordant sign of dysregulation. Permutation null p < 0.01 (10,000 random L-R subsets).\n\nH2 (secondary — Hong-MR-gene enrichment): The shared L-R pairs are enriched (q < 0.05 hypergeometric) for ligands or receptors whose genes appear on the Hong 2025 MR-implicated immune-gene list (FDFT1, CTSB, ARSA, HLA-DQA1, and 24 others reported in Hong 2025 Table 2).\n\nH3 (secondary — peripheral-foreshadow correlation): A \"peripheral CD8+T signature score\" computed from Hong 2025 PBMC data using the 28 MR-implicated immune genes rank-correlates (Spearman ρ > 0.3, nominal p < 0.05) with the v1 `parkinson-tcell-foothold` per-donor brain T-cell-module score at the disease-stage level (early vs. late PD).\n\nH4 (secondary — direction-specific dropout): Inferring CD8+T → monocyte (PBMC) and microglia → CD8+T (brain) — the reverse-direction pairs — should NOT show the same H1 enrichment (control: PD signals are in the CD8+T → myeloid direction, not the reverse).\n\nH5 (secondary — method-consensus robustness): Repeat the L-R inference with three methods (CellPhoneDB, NATMI, scTalk) and confirm the H1 shared-pair count is within ±2 of the LIANA consensus result. (LIANA-py provides this consensus natively.)\n\nNull payoff. If peripheral and brain L-R signatures diverge (H1: 0-3 shared pairs), peripheral-blood-as-PD-biomarker biology is decoupled from the brain-immune axis — informative, with strong implications for blood-test development.", "report": null, "page": "/campaign/runs/parkinson-cellchat-immune-axis-r2/" }, { "run": "parkinson-cellchat-immune-axis-r3", "topic": "parkinson-cellchat-immune-axis", "attempt": 3, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 2, "partial": 0, "not_supported": 1, "inconclusive": 0, "not_testable": 1, "not_run": 1 }, "evaluated": 4, "stop_reason": null, "question": "Do LIANA-py (CellChat-equivalent consensus) inferred ligand- receptor (L-R) communications between Hong 2025 peripheral-blood CD8+ T cells (GEO GSE223138) and Fullard brain microglia mirror — at the L-R pair level — the brain-internal CD8+T → microglia communications inferred by Ma 2025 (PMC12322057), thus identifying a \"blood-as-foreshadow-of-brain\" set of peripheral biomarker ligands?\n\nHong 2025 (npj Parkinson's 11) reported 28 immune-cell-specific PD risk genes from OneK1K + IPDGC + GSE223138 multi-omic integration. Ma 2025 (Nat Commun 16) characterised brain-side CD8+ T-cell × microglia communications via SN scRNA + scTCR-seq. The two cell-type-communication signatures have not been compared at the L-R pair level. If peripheral L-Rs echo brain L-Rs, the peripheral signal could be used as a foreshadow blood biomarker; if they diverge, peripheral-blood-as-PD-biomarker biology is decoupled from the brain-immune axis. Either outcome is informative with strong biomarker implications.\n\nHypotheses (pre-registered):\n\nH1 (primary — L-R pair concordance): ≥ 5 ligand-receptor pairs are inferred as PD-up-regulated in BOTH the peripheral CD8+T → blood-monocyte direction (Hong 2025) AND the brain CD8+T → microglia direction (Ma 2025), with concordant sign of dysregulation. Permutation null p < 0.01 (10,000 random L-R subsets).\n\nH2 (secondary — Hong-MR-gene enrichment): The shared L-R pairs are enriched (q < 0.05 hypergeometric) for ligands or receptors whose genes appear on the Hong 2025 MR-implicated immune-gene list (FDFT1, CTSB, ARSA, HLA-DQA1, and 24 others reported in Hong 2025 Table 2).\n\nH3 (secondary — peripheral-foreshadow correlation): A \"peripheral CD8+T signature score\" computed from Hong 2025 PBMC data using the 28 MR-implicated immune genes rank-correlates (Spearman ρ > 0.3, nominal p < 0.05) with the v1 `parkinson-tcell-foothold` per-donor brain T-cell-module score at the disease-stage level (early vs. late PD).\n\nH4 (secondary — direction-specific dropout): Inferring CD8+T → monocyte (PBMC) and microglia → CD8+T (brain) — the reverse-direction pairs — should NOT show the same H1 enrichment (control: PD signals are in the CD8+T → myeloid direction, not the reverse).\n\nH5 (secondary — method-consensus robustness): Repeat the L-R inference with three methods (CellPhoneDB, NATMI, scTalk) and confirm the H1 shared-pair count is within ±2 of the LIANA consensus result. (LIANA-py provides this consensus natively.)\n\nNull payoff. If peripheral and brain L-R signatures diverge (H1: 0-3 shared pairs), peripheral-blood-as-PD-biomarker biology is decoupled from the brain-immune axis — informative, with strong implications for blood-test development.", "report": { "kind": "pdf", "url": "/assets/reports/parkinson-cellchat-immune-axis-r3.pdf", "bytes": 280249, "title": "Nine concordant peripheral-brain CD8+ T cell ligand-receptor pairs in Parkinson's disease: a blood-as-foreshadow analysis", "main_conclusion": "Nine PD-upregulated CD8+ T to myeloid L-R pairs were shared between PBMC and non-SN brain axes (observed 9 vs null mean 1.39, Delta = 7.607, p < 1/10,000; H1 supported). MR-gene enrichment was not significant (H2: OR = 2.56, p = 0.346, underpowered). Direction-specificity was borderline (H4: reverse p = 0.053). H5 robustness was met, but the methods were not truly independent, and PD specificity remains unestablished.", "thumb": "/assets/img/thumbs/parkinson-cellchat-immune-axis-r3.webp" }, "page": "/campaign/runs/parkinson-cellchat-immune-axis-r3/" }, { "run": "parkinson-celloracle-ko-r1", "topic": "parkinson-celloracle-ko", "attempt": 1, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 6, "counts": { "supported": 3, "partial": 0, "not_supported": 2, "inconclusive": 0, "not_testable": 0, "not_run": 1 }, "evaluated": 5, "stop_reason": null, "question": "Does CellOracle's in-silico knock-out (KO) of the top PD-risk genes from the Kim 2024 multi-ancestry GWAS — LRRK2, GBA, SNCA, VPS13C, GAK, MAPT, BST1, CTSB, DDC, SCARB2 (10 genes ranked by fine-mapping posterior probability) — in Fullard 2024 SNpc dopaminergic neurons produce predicted cell-state vector shifts that align with the observed PD-vs-control DA state shift?\n\nThe CellOracle framework (Kamimoto et al. 2023 Nature) lets one combine a scATAC-derived base-GRN with snRNA-seq to predict counter-factual cell-state shifts in response to TF / target perturbation. The mechanistic-content claim — that the inferred GRN encodes enough biology to predict disease-state shifts — has not been formally tested in human PD with public data.\n\nHypotheses (pre-registered):\n\nH1 (primary — KO-vs-disease alignment): The mean cosine similarity between the CellOracle-predicted post-KO state-shift vector (averaged across the 10 GWAS genes) and the observed PD-minus-control SNpc DA centroid shift is > 0.30, with a permutation-null p < 0.01 (10,000 random gene-label permutations).\n\nH2 (secondary — fine-mapping PIP correlation): Per-gene KO cosine alignment with PD-minus-control shift correlates positively (Spearman ρ > 0.4, p < 0.05) with the gene's posterior probability of causal association from Kim 2024 fine-mapping (Supp Table S5).\n\nH3 (secondary — LRRK2-KO target enrichment): The CellOracle-predicted top-100 up-regulated targets of LRRK2-KO are enriched (q < 0.05 by hypergeometric) for the Hemedan 2024 Boolean-PD \"prodromal-cluster\" regulatory- module gene list.\n\nH4 (secondary — SNCA protective + HBB null control): SNCA-KO simulation predicts a *protective* shift (cosine with PD-vs-ctrl vector > 0 with permutation p < 0.05), consistent with α-syn loss-of-function reducing aggregation; control gene HBB (red-blood-cell marker, irrelevant) predicts a null shift (|cosine| < 0.1).\n\nH5 (secondary — base-GRN robustness): Repeat the analysis with the published CellOracle midbrain reference base-GRN (Zenodo 7755357) vs. the Adams-2024- derived ATAC base-GRN. The H1 cosine-similarity changes by ≤ 0.1 between the two base-GRN sources (network-source sensitivity test).\n\nNull payoff. A null cosine alignment (H1 ≤ 0.15) means GWAS- implicated genes do not in-silico recapitulate PD cellular phenotypes — sharpening the long-standing field critique that risk-variant biology is not cell-autonomous and may be glia- mediated.", "report": { "kind": "pdf", "url": "/assets/reports/parkinson-celloracle-ko-r1.pdf", "bytes": 425566, "title": "In-silico knock-out of ten Parkinson's disease GWAS genes produces sub-threshold but above-null dopaminergic cell-state shifts: a validation protocol for CellOracle-based GRN perturbation", "main_conclusion": "H1 was not supported. The mean GWAS-gene KO cosine with the PD disease vector was 0.025 (95% CI -0.067 to 0.114), below the 0.30 threshold, although it beat the permutation null (Delta = 0.159, p = 0.001). H2 (PIP correlation, rho = 0.042) and H5 (|Delta| = 0.118) were not supported. H3 (LRRK2-KO prodromal enrichment, FE = 4.7x, q = 0.007), H4a (SNCA, +0.235) and H4b (HBB null) were supported. These results come from a stratification approximation (CellOracle ATAC GRN unavailable) run in the DMNX_Neu fallback tissue.", "thumb": "/assets/img/thumbs/parkinson-celloracle-ko-r1.webp" }, "page": "/campaign/runs/parkinson-celloracle-ko-r1/" }, { "run": "parkinson-da-vulnerability-r1", "topic": "parkinson-da-vulnerability", "attempt": 1, "disease": "PD", "census_class": "EARLY", "final_status": "halted", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Kamath, Macosko et al. (Nat Neurosci 25:588, 2022; DOI 10.1038/s41593-022-01061-1) reported that a single AGTR1+ ventral-tier substantia-nigra pars compacta (SNpc) dopaminergic (DA) neuronal subtype is selectively lost in Parkinson's disease (PD), and that residual AGTR1+ DA cells upregulate TP53 and NR2F2 target programs. The finding was derived from 10 PD/LBD + 8 control donors (387,483 nuclei; 22,048 DA-restricted nuclei) and spatially anchored by Slide-seqV2 to the ventral tier of SNpc. It is the most-cited cellular finding in PD single-cell biology, but has never been formally replicated against three subsequent *independent* SNpc snRNA-seq atlases under a single re-analysis pipeline.\n\nThis OmiCairn run formally tests whether the AGTR1+ vulnerable DA subtype is recoverable, and whether it shows the same PD-specific depletion + TP53/NR2F2 upregulation, in:\n\n- Wang et al. (Sci Adv 10:eadi8287, 2024; GEO GSE184950) — 23 idiopathic PD + 9 control / 315,867 SNpc nuclei.\n\n- Adams L., Song M.K. et al. (Nat Aging 4:364, 2024; GEO GSE193688 + dbGaP phs002819.v1.p1) — 9 young + 8 aged + 14 PD / 69,289 SN nuclei, 10x Multiome.\n\n- Fullard / N M et al. (Sci Data 11:1274, 2024; CELLxGENE collection d5d0df8f-4eee-49d8-a221-a288f50a1590) — 100 donors (75 PD + 25 ctrl), 2,096,155 nuclei across 5 regions; this proposal uses the SNpc slab only.\n\n- Martirosyan et al. (Mol Neurodegen 19:7, 2024; GEO GSE243639) — 15 PD + 14 ctrl / ~84,000 SNpc nuclei; held out for blind external validation in H3.\n\nHypotheses (pre-registered):\n\nH1 (primary — vulnerable-subtype depletion replicates): The Kamath AGTR1+ DA subtype (defined by the published Kamath top-30 marker panel — AGTR1, SLC6A3, TH, KCNJ6, etc.; full list lifted from GSE178265 supplementary table) recovers as a contiguous cluster in each of Wang (A3), Adams (A4), and Fullard SNpc (A8) atlases at our chosen clustering resolution. The log-fold-change of (AGTR1+ DA cells / total DA cells) PD vs. matched-age control is negative (depletion) in all 3 replication atlases, with FDR < 0.05 (Benjamini–Hochberg). Effect-size requirement: the magnitude of depletion is within ±50% of the Kamath reference (≈3-fold loss → our log2FC ∈ [-2.6, -0.95]).\n\nH2 (secondary — TP53 / NR2F2 program upregulation): Within retained AGTR1+ DA cells, the TP53 target module and NR2F2 target module (lifted directly from Hallmark TP53 signaling + the Kamath-cited NR2F2 ChIP-seq targets) show positive PD-vs-control fold-change at FDR < 0.05 in ≥ 2 of 3 replication atlases.\n\nH3 (secondary — blind-replication out-of-cohort): A \"consensus AGTR1+ marker panel\" (intersection of the marker-gene sets across Kamath + Wang + Adams + Fullard SNpc, re-derived from our own DE analysis) scored as a per-donor pseudobulk signature stratifies the Martirosyan A5 PD vs. control donors with ROC AUC ≥ 0.75 (95% bootstrap CI excluding 0.5).\n\nH4 (secondary — cluster-resolution sensitivity): We report the sensitivity of H1 to clustering resolution (Leiden / Louvain γ ∈ {0.3, 0.5, 0.8, 1.0, 1.5, 2.0}). H1's sign and FDR pass are stable across at least 4 of the 6 resolutions. (A finding that AGTR1+ depletion is observed only at γ ≈ Kamath's published resolution would itself be biologically informative and must be reported.)\n\nH5 (secondary — spatial vs. proportional vs. expression decomposition): We decompose PD-vs-control DA loss into: (a) total nuclei recovered per donor (sequencing depth / dissection artifact) (b) DA-cluster fraction within total nuclei (c) AGTR1+ subcluster fraction within DA (d) per-cell AGTR1 expression within AGTR1+ DA and report each of (a)–(d) separately. Honest replication of Kamath requires the effect to be predominantly in (c), not in (a) or (b). A dominant (a) effect would indicate the original finding is partly a dissection or doublet artifact.\n\nA negative or partial-replication result is fully informative. This OmiCairn run must report cluster-merging, marker-gene-loss, or resolution-instability findings as the headline rather than tune clustering until H1 passes.", "report": null, "page": "/campaign/runs/parkinson-da-vulnerability-r1/" }, { "run": "parkinson-da-vulnerability-r2", "topic": "parkinson-da-vulnerability", "attempt": 2, "disease": "PD", "census_class": "EARLY", "final_status": "halted", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Kamath, Macosko et al. (Nat Neurosci 25:588, 2022; DOI 10.1038/s41593-022-01061-1) reported that a single AGTR1+ ventral-tier substantia-nigra pars compacta (SNpc) dopaminergic (DA) neuronal subtype is selectively lost in Parkinson's disease (PD), and that residual AGTR1+ DA cells upregulate TP53 and NR2F2 target programs. The finding was derived from 10 PD/LBD + 8 control donors (387,483 nuclei; 22,048 DA-restricted nuclei) and spatially anchored by Slide-seqV2 to the ventral tier of SNpc. It is the most-cited cellular finding in PD single-cell biology, but has never been formally replicated against three subsequent *independent* SNpc snRNA-seq atlases under a single re-analysis pipeline.\n\nThis OmiCairn run formally tests whether the AGTR1+ vulnerable DA subtype is recoverable, and whether it shows the same PD-specific depletion + TP53/NR2F2 upregulation, in:\n\n- Wang et al. (Sci Adv 10:eadi8287, 2024; GEO GSE184950) — 23 idiopathic PD + 9 control / 315,867 SNpc nuclei.\n\n- Adams L., Song M.K. et al. (Nat Aging 4:364, 2024; GEO GSE193688 + dbGaP phs002819.v1.p1) — 9 young + 8 aged + 14 PD / 69,289 SN nuclei, 10x Multiome.\n\n- Fullard / N M et al. (Sci Data 11:1274, 2024; CELLxGENE collection d5d0df8f-4eee-49d8-a221-a288f50a1590) — 100 donors (75 PD + 25 ctrl), 2,096,155 nuclei across 5 regions; this proposal uses the SNpc slab only.\n\n- Martirosyan et al. (Mol Neurodegen 19:7, 2024; GEO GSE243639) — 15 PD + 14 ctrl / ~84,000 SNpc nuclei; held out for blind external validation in H3.\n\nHypotheses (pre-registered):\n\nH1 (primary — vulnerable-subtype depletion replicates): The Kamath AGTR1+ DA subtype (defined by the published Kamath top-30 marker panel — AGTR1, SLC6A3, TH, KCNJ6, etc.; full list lifted from GSE178265 supplementary table) recovers as a contiguous cluster in each of Wang (A3), Adams (A4), and Fullard SNpc (A8) atlases at our chosen clustering resolution. The log-fold-change of (AGTR1+ DA cells / total DA cells) PD vs. matched-age control is negative (depletion) in all 3 replication atlases, with FDR < 0.05 (Benjamini–Hochberg). Effect-size requirement: the magnitude of depletion is within ±50% of the Kamath reference (≈3-fold loss → our log2FC ∈ [-2.6, -0.95]).\n\nH2 (secondary — TP53 / NR2F2 program upregulation): Within retained AGTR1+ DA cells, the TP53 target module and NR2F2 target module (lifted directly from Hallmark TP53 signaling + the Kamath-cited NR2F2 ChIP-seq targets) show positive PD-vs-control fold-change at FDR < 0.05 in ≥ 2 of 3 replication atlases.\n\nH3 (secondary — blind-replication out-of-cohort): A \"consensus AGTR1+ marker panel\" (intersection of the marker-gene sets across Kamath + Wang + Adams + Fullard SNpc, re-derived from our own DE analysis) scored as a per-donor pseudobulk signature stratifies the Martirosyan A5 PD vs. control donors with ROC AUC ≥ 0.75 (95% bootstrap CI excluding 0.5).\n\nH4 (secondary — cluster-resolution sensitivity): We report the sensitivity of H1 to clustering resolution (Leiden / Louvain γ ∈ {0.3, 0.5, 0.8, 1.0, 1.5, 2.0}). H1's sign and FDR pass are stable across at least 4 of the 6 resolutions. (A finding that AGTR1+ depletion is observed only at γ ≈ Kamath's published resolution would itself be biologically informative and must be reported.)\n\nH5 (secondary — spatial vs. proportional vs. expression decomposition): We decompose PD-vs-control DA loss into: (a) total nuclei recovered per donor (sequencing depth / dissection artifact) (b) DA-cluster fraction within total nuclei (c) AGTR1+ subcluster fraction within DA (d) per-cell AGTR1 expression within AGTR1+ DA and report each of (a)–(d) separately. Honest replication of Kamath requires the effect to be predominantly in (c), not in (a) or (b). A dominant (a) effect would indicate the original finding is partly a dissection or doublet artifact.\n\nA negative or partial-replication result is fully informative. This OmiCairn run must report cluster-merging, marker-gene-loss, or resolution-instability findings as the headline rather than tune clustering until H1 passes.", "report": null, "page": "/campaign/runs/parkinson-da-vulnerability-r2/" }, { "run": "parkinson-da-vulnerability-r3", "topic": "parkinson-da-vulnerability", "attempt": 3, "disease": "PD", "census_class": "EARLY", "final_status": "halted", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Kamath, Macosko et al. (Nat Neurosci 25:588, 2022; DOI 10.1038/s41593-022-01061-1) reported that a single AGTR1+ ventral-tier substantia-nigra pars compacta (SNpc) dopaminergic (DA) neuronal subtype is selectively lost in Parkinson's disease (PD), and that residual AGTR1+ DA cells upregulate TP53 and NR2F2 target programs. The finding was derived from 10 PD/LBD + 8 control donors (387,483 nuclei; 22,048 DA-restricted nuclei) and spatially anchored by Slide-seqV2 to the ventral tier of SNpc. It is the most-cited cellular finding in PD single-cell biology, but has never been formally replicated against three subsequent *independent* SNpc snRNA-seq atlases under a single re-analysis pipeline.\n\nThis OmiCairn run formally tests whether the AGTR1+ vulnerable DA subtype is recoverable, and whether it shows the same PD-specific depletion + TP53/NR2F2 upregulation, in:\n\n- Wang et al. (Sci Adv 10:eadi8287, 2024; GEO GSE184950) — 23 idiopathic PD + 9 control / 315,867 SNpc nuclei.\n\n- Adams L., Song M.K. et al. (Nat Aging 4:364, 2024; GEO GSE193688 + dbGaP phs002819.v1.p1) — 9 young + 8 aged + 14 PD / 69,289 SN nuclei, 10x Multiome.\n\n- Fullard / N M et al. (Sci Data 11:1274, 2024; CELLxGENE collection d5d0df8f-4eee-49d8-a221-a288f50a1590) — 100 donors (75 PD + 25 ctrl), 2,096,155 nuclei across 5 regions; this proposal uses the SNpc slab only.\n\n- Martirosyan et al. (Mol Neurodegen 19:7, 2024; GEO GSE243639) — 15 PD + 14 ctrl / ~84,000 SNpc nuclei; held out for blind external validation in H3.\n\nHypotheses (pre-registered):\n\nH1 (primary — vulnerable-subtype depletion replicates): The Kamath AGTR1+ DA subtype (defined by the published Kamath top-30 marker panel — AGTR1, SLC6A3, TH, KCNJ6, etc.; full list lifted from GSE178265 supplementary table) recovers as a contiguous cluster in each of Wang (A3), Adams (A4), and Fullard SNpc (A8) atlases at our chosen clustering resolution. The log-fold-change of (AGTR1+ DA cells / total DA cells) PD vs. matched-age control is negative (depletion) in all 3 replication atlases, with FDR < 0.05 (Benjamini–Hochberg). Effect-size requirement: the magnitude of depletion is within ±50% of the Kamath reference (≈3-fold loss → our log2FC ∈ [-2.6, -0.95]).\n\nH2 (secondary — TP53 / NR2F2 program upregulation): Within retained AGTR1+ DA cells, the TP53 target module and NR2F2 target module (lifted directly from Hallmark TP53 signaling + the Kamath-cited NR2F2 ChIP-seq targets) show positive PD-vs-control fold-change at FDR < 0.05 in ≥ 2 of 3 replication atlases.\n\nH3 (secondary — blind-replication out-of-cohort): A \"consensus AGTR1+ marker panel\" (intersection of the marker-gene sets across Kamath + Wang + Adams + Fullard SNpc, re-derived from our own DE analysis) scored as a per-donor pseudobulk signature stratifies the Martirosyan A5 PD vs. control donors with ROC AUC ≥ 0.75 (95% bootstrap CI excluding 0.5).\n\nH4 (secondary — cluster-resolution sensitivity): We report the sensitivity of H1 to clustering resolution (Leiden / Louvain γ ∈ {0.3, 0.5, 0.8, 1.0, 1.5, 2.0}). H1's sign and FDR pass are stable across at least 4 of the 6 resolutions. (A finding that AGTR1+ depletion is observed only at γ ≈ Kamath's published resolution would itself be biologically informative and must be reported.)\n\nH5 (secondary — spatial vs. proportional vs. expression decomposition): We decompose PD-vs-control DA loss into: (a) total nuclei recovered per donor (sequencing depth / dissection artifact) (b) DA-cluster fraction within total nuclei (c) AGTR1+ subcluster fraction within DA (d) per-cell AGTR1 expression within AGTR1+ DA and report each of (a)–(d) separately. Honest replication of Kamath requires the effect to be predominantly in (c), not in (a) or (b). A dominant (a) effect would indicate the original finding is partly a dissection or doublet artifact.\n\nA negative or partial-replication result is fully informative. This OmiCairn run must report cluster-merging, marker-gene-loss, or resolution-instability findings as the headline rather than tune clustering until H1 passes.", "report": null, "page": "/campaign/runs/parkinson-da-vulnerability-r3/" }, { "run": "parkinson-da-vulnerability-r4", "topic": "parkinson-da-vulnerability", "attempt": 4, "disease": "PD", "census_class": "PARTIAL", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 4, "not_testable": 0, "not_run": 1 }, "evaluated": 4, "stop_reason": null, "question": "Kamath, Macosko et al. (Nat Neurosci 25:588, 2022; DOI 10.1038/s41593-022-01061-1) reported that a single AGTR1+ ventral-tier substantia-nigra pars compacta (SNpc) dopaminergic (DA) neuronal subtype is selectively lost in Parkinson's disease (PD), and that residual AGTR1+ DA cells upregulate TP53 and NR2F2 target programs. The finding was derived from 10 PD/LBD + 8 control donors (387,483 nuclei; 22,048 DA-restricted nuclei) and spatially anchored by Slide-seqV2 to the ventral tier of SNpc. It is the most-cited cellular finding in PD single-cell biology, but has never been formally replicated against three subsequent *independent* SNpc snRNA-seq atlases under a single re-analysis pipeline.\n\nThis OmiCairn run formally tests whether the AGTR1+ vulnerable DA subtype is recoverable, and whether it shows the same PD-specific depletion + TP53/NR2F2 upregulation, in:\n\n- Wang et al. (Sci Adv 10:eadi8287, 2024; GEO GSE184950) — 23 idiopathic PD + 9 control / 315,867 SNpc nuclei.\n\n- Adams L., Song M.K. et al. (Nat Aging 4:364, 2024; GEO GSE193688 + dbGaP phs002819.v1.p1) — 9 young + 8 aged + 14 PD / 69,289 SN nuclei, 10x Multiome.\n\n- Fullard / N M et al. (Sci Data 11:1274, 2024; CELLxGENE collection d5d0df8f-4eee-49d8-a221-a288f50a1590) — 100 donors (75 PD + 25 ctrl), 2,096,155 nuclei across 5 regions; this proposal uses the SNpc slab only.\n\n- Martirosyan et al. (Mol Neurodegen 19:7, 2024; GEO GSE243639) — 15 PD + 14 ctrl / ~84,000 SNpc nuclei; held out for blind external validation in H3.\n\nHypotheses (pre-registered):\n\nH1 (primary — vulnerable-subtype depletion replicates): The Kamath AGTR1+ DA subtype (defined by the published Kamath top-30 marker panel — AGTR1, SLC6A3, TH, KCNJ6, etc.; full list lifted from GSE178265 supplementary table) recovers as a contiguous cluster in each of Wang (A3), Adams (A4), and Fullard SNpc (A8) atlases at our chosen clustering resolution. The log-fold-change of (AGTR1+ DA cells / total DA cells) PD vs. matched-age control is negative (depletion) in all 3 replication atlases, with FDR < 0.05 (Benjamini–Hochberg). Effect-size requirement: the magnitude of depletion is within ±50% of the Kamath reference (≈3-fold loss → our log2FC ∈ [-2.6, -0.95]).\n\nH2 (secondary — TP53 / NR2F2 program upregulation): Within retained AGTR1+ DA cells, the TP53 target module and NR2F2 target module (lifted directly from Hallmark TP53 signaling + the Kamath-cited NR2F2 ChIP-seq targets) show positive PD-vs-control fold-change at FDR < 0.05 in ≥ 2 of 3 replication atlases.\n\nH3 (secondary — blind-replication out-of-cohort): A \"consensus AGTR1+ marker panel\" (intersection of the marker-gene sets across Kamath + Wang + Adams + Fullard SNpc, re-derived from our own DE analysis) scored as a per-donor pseudobulk signature stratifies the Martirosyan A5 PD vs. control donors with ROC AUC ≥ 0.75 (95% bootstrap CI excluding 0.5).\n\nH4 (secondary — cluster-resolution sensitivity): We report the sensitivity of H1 to clustering resolution (Leiden / Louvain γ ∈ {0.3, 0.5, 0.8, 1.0, 1.5, 2.0}). H1's sign and FDR pass are stable across at least 4 of the 6 resolutions. (A finding that AGTR1+ depletion is observed only at γ ≈ Kamath's published resolution would itself be biologically informative and must be reported.)\n\nH5 (secondary — spatial vs. proportional vs. expression decomposition): We decompose PD-vs-control DA loss into: (a) total nuclei recovered per donor (sequencing depth / dissection artifact) (b) DA-cluster fraction within total nuclei (c) AGTR1+ subcluster fraction within DA (d) per-cell AGTR1 expression within AGTR1+ DA and report each of (a)–(d) separately. Honest replication of Kamath requires the effect to be predominantly in (c), not in (a) or (b). A dominant (a) effect would indicate the original finding is partly a dissection or doublet artifact.\n\nA negative or partial-replication result is fully informative. This OmiCairn run must report cluster-merging, marker-gene-loss, or resolution-instability findings as the headline rather than tune clustering until H1 passes.", "report": null, "page": "/campaign/runs/parkinson-da-vulnerability-r4/" }, { "run": "parkinson-da-vulnerability-r5", "topic": "parkinson-da-vulnerability", "attempt": 5, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 0, "not_supported": 2, "inconclusive": 2, "not_testable": 1, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Kamath, Macosko et al. (Nat Neurosci 25:588, 2022; DOI 10.1038/s41593-022-01061-1) reported that a single AGTR1+ ventral-tier substantia-nigra pars compacta (SNpc) dopaminergic (DA) neuronal subtype is selectively lost in Parkinson's disease (PD), and that residual AGTR1+ DA cells upregulate TP53 and NR2F2 target programs. The finding was derived from 10 PD/LBD + 8 control donors (387,483 nuclei; 22,048 DA-restricted nuclei) and spatially anchored by Slide-seqV2 to the ventral tier of SNpc. It is the most-cited cellular finding in PD single-cell biology, but has never been formally replicated against three subsequent *independent* SNpc snRNA-seq atlases under a single re-analysis pipeline.\n\nThis OmiCairn run formally tests whether the AGTR1+ vulnerable DA subtype is recoverable, and whether it shows the same PD-specific depletion + TP53/NR2F2 upregulation, in:\n\n- Wang et al. (Sci Adv 10:eadi8287, 2024; GEO GSE184950) — 23 idiopathic PD + 9 control / 315,867 SNpc nuclei.\n\n- Adams L., Song M.K. et al. (Nat Aging 4:364, 2024; GEO GSE193688 + dbGaP phs002819.v1.p1) — 9 young + 8 aged + 14 PD / 69,289 SN nuclei, 10x Multiome.\n\n- Fullard / N M et al. (Sci Data 11:1274, 2024; CELLxGENE collection d5d0df8f-4eee-49d8-a221-a288f50a1590) — 100 donors (75 PD + 25 ctrl), 2,096,155 nuclei across 5 regions; this proposal uses the SNpc slab only.\n\n- Martirosyan et al. (Mol Neurodegen 19:7, 2024; GEO GSE243639) — 15 PD + 14 ctrl / ~84,000 SNpc nuclei; held out for blind external validation in H3.\n\nHypotheses (pre-registered):\n\nH1 (primary — vulnerable-subtype depletion replicates): The Kamath AGTR1+ DA subtype (defined by the published Kamath top-30 marker panel — AGTR1, SLC6A3, TH, KCNJ6, etc.; full list lifted from GSE178265 supplementary table) recovers as a contiguous cluster in each of Wang (A3), Adams (A4), and Fullard SNpc (A8) atlases at our chosen clustering resolution. The log-fold-change of (AGTR1+ DA cells / total DA cells) PD vs. matched-age control is negative (depletion) in all 3 replication atlases, with FDR < 0.05 (Benjamini–Hochberg). Effect-size requirement: the magnitude of depletion is within ±50% of the Kamath reference (≈3-fold loss → our log2FC ∈ [-2.6, -0.95]).\n\nH2 (secondary — TP53 / NR2F2 program upregulation): Within retained AGTR1+ DA cells, the TP53 target module and NR2F2 target module (lifted directly from Hallmark TP53 signaling + the Kamath-cited NR2F2 ChIP-seq targets) show positive PD-vs-control fold-change at FDR < 0.05 in ≥ 2 of 3 replication atlases.\n\nH3 (secondary — blind-replication out-of-cohort): A \"consensus AGTR1+ marker panel\" (intersection of the marker-gene sets across Kamath + Wang + Adams + Fullard SNpc, re-derived from our own DE analysis) scored as a per-donor pseudobulk signature stratifies the Martirosyan A5 PD vs. control donors with ROC AUC ≥ 0.75 (95% bootstrap CI excluding 0.5).\n\nH4 (secondary — cluster-resolution sensitivity): We report the sensitivity of H1 to clustering resolution (Leiden / Louvain γ ∈ {0.3, 0.5, 0.8, 1.0, 1.5, 2.0}). H1's sign and FDR pass are stable across at least 4 of the 6 resolutions. (A finding that AGTR1+ depletion is observed only at γ ≈ Kamath's published resolution would itself be biologically informative and must be reported.)\n\nH5 (secondary — spatial vs. proportional vs. expression decomposition): We decompose PD-vs-control DA loss into: (a) total nuclei recovered per donor (sequencing depth / dissection artifact) (b) DA-cluster fraction within total nuclei (c) AGTR1+ subcluster fraction within DA (d) per-cell AGTR1 expression within AGTR1+ DA and report each of (a)–(d) separately. Honest replication of Kamath requires the effect to be predominantly in (c), not in (a) or (b). A dominant (a) effect would indicate the original finding is partly a dissection or doublet artifact.\n\nA negative or partial-replication result is fully informative. This OmiCairn run must report cluster-merging, marker-gene-loss, or resolution-instability findings as the headline rather than tune clustering until H1 passes.", "report": { "kind": "pdf", "url": "/assets/reports/parkinson-da-vulnerability-r5.pdf", "bytes": 276764, "title": "Pre-registered null replication of AGTR1+ dopaminergic vulnerability in Parkinson's disease across independent SNpc atlases", "main_conclusion": "Pre-registered null: the Kamath AGTR1+ DA subtype was not recoverable in either replication atlas (replication 0/2, Wilson CI 0-65.8%, perm p=1.0; Wang log2FC -0.263 FDR 0.902; Adams +1.000 FDR 0.605), H2 inconclusive, H3 consensus panel failed blind validation (AUC 0.419), H4 0/6 stable resolutions. The authors attribute the failure mainly to low DA-neuron yield (52 and 27 cells vs 22,048).", "thumb": "/assets/img/thumbs/parkinson-da-vulnerability-r5.webp" }, "page": "/campaign/runs/parkinson-da-vulnerability-r5/" }, { "run": "parkinson-foundation-embedding-r1", "topic": "parkinson-foundation-embedding", "attempt": 1, "disease": "PD", "census_class": "PARTIAL", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 5 }, "evaluated": 0, "stop_reason": null, "question": "Do zero-shot Geneformer (95M-cell variant) and scGPT embeddings of Fullard 2024 SNpc dopaminergic-neuron nuclei separate PD from control donors better than a baseline log-CPM-PCA-LR pipeline at the donor level — and which foundation model wins?\n\nCui et al. 2024 (scGPT, Nat Methods 21:1470) and Theodoris et al. 2023+2024 (Geneformer, Nature 618:616 + 95M update) made strong claims about disease-stratification utility of foundation-model embeddings on cell atlases. Boiarsky 2024 and Kedzierska 2024 have begun to publish benchmarks showing zero-shot foundation models are often *worse* than baseline log-CPM-PCA-LR in disease contexts. PD has not been placed on that benchmark map. This OmiCairn run does so.\n\nHypotheses (pre-registered):\n\nH1 (primary — donor-level classification, foundation model vs. baseline): Donor-level leave-one-out (LOO) AUC for PD-vs-control classification using donor-mean foundation-model embedding + logistic regression is > 0.75 for at least one of Geneformer or scGPT, AND that AUC is > the log-CPM-PCA-LR baseline AUC by ΔAUC ≥ 0.05, with DeLong test p < 0.05.\n\nH2 (secondary — AGTR1+ subcluster recovery without supervision): Embedding-derived nearest-neighbour graph (Leiden γ = 0.6) in PD donors recovers an AGTR1+ vs. AGTR1- subcluster partition with Adjusted Rand Index ≥ 0.4 against the v1 `parkinson-da-vulnerability` workspace labels.\n\nH3 (secondary — transfer-learning benefit): Geneformer fine-tuned for 1 epoch on Kamath A1 PD-vs-ctrl labels (GSE178265) generalises to Wang (GSE184950) and Adams (GSE193688) atlases with held-out AUC ≥ 0.7 in each, vs. ≤ 0.6 for the same architecture trained from scratch on the same fine-tune set.\n\nH4 (secondary — embedding-space robustness): Repeat H1 with three subsampling levels (5K / 25K / 50K nuclei per donor average); the ΔAUC vs. baseline must be stable (within ±0.05) across the three subsamples, otherwise the foundation-model advantage is sample-size dependent and that must be reported.\n\nH5 (secondary — cross-atlas zero-shot generalisation): Foundation-model embeddings trained on Fullard alone achieve PD-vs-ctrl AUC ≥ 0.65 when scored zero-shot on Wang and Adams (no fine-tuning). Below this floor, the embedding is Fullard-specific rather than generalising.\n\nNull payoff. If neither Geneformer nor scGPT beats log-CPM-PCA-LR (ΔAUC < 0.05), the field's optimism about single-cell foundation models for neurodegenerative disease is overstated for PD — matching the Boiarsky 2024 + Kedzierska 2024 findings in other diseases, and placing PD on the benchmark map.", "report": null, "page": "/campaign/runs/parkinson-foundation-embedding-r1/" }, { "run": "parkinson-foundation-embedding-r2", "topic": "parkinson-foundation-embedding", "attempt": 2, "disease": "PD", "census_class": "PARTIAL", "final_status": "in_progress", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Do zero-shot Geneformer (95M-cell variant) and scGPT embeddings of Fullard 2024 SNpc dopaminergic-neuron nuclei separate PD from control donors better than a baseline log-CPM-PCA-LR pipeline at the donor level — and which foundation model wins?\n\nCui et al. 2024 (scGPT, Nat Methods 21:1470) and Theodoris et al. 2023+2024 (Geneformer, Nature 618:616 + 95M update) made strong claims about disease-stratification utility of foundation-model embeddings on cell atlases. Boiarsky 2024 and Kedzierska 2024 have begun to publish benchmarks showing zero-shot foundation models are often *worse* than baseline log-CPM-PCA-LR in disease contexts. PD has not been placed on that benchmark map. This OmiCairn run does so.\n\nHypotheses (pre-registered):\n\nH1 (primary — donor-level classification, foundation model vs. baseline): Donor-level leave-one-out (LOO) AUC for PD-vs-control classification using donor-mean foundation-model embedding + logistic regression is > 0.75 for at least one of Geneformer or scGPT, AND that AUC is > the log-CPM-PCA-LR baseline AUC by ΔAUC ≥ 0.05, with DeLong test p < 0.05.\n\nH2 (secondary — AGTR1+ subcluster recovery without supervision): Embedding-derived nearest-neighbour graph (Leiden γ = 0.6) in PD donors recovers an AGTR1+ vs. AGTR1- subcluster partition with Adjusted Rand Index ≥ 0.4 against the v1 `parkinson-da-vulnerability` workspace labels.\n\nH3 (secondary — transfer-learning benefit): Geneformer fine-tuned for 1 epoch on Kamath A1 PD-vs-ctrl labels (GSE178265) generalises to Wang (GSE184950) and Adams (GSE193688) atlases with held-out AUC ≥ 0.7 in each, vs. ≤ 0.6 for the same architecture trained from scratch on the same fine-tune set.\n\nH4 (secondary — embedding-space robustness): Repeat H1 with three subsampling levels (5K / 25K / 50K nuclei per donor average); the ΔAUC vs. baseline must be stable (within ±0.05) across the three subsamples, otherwise the foundation-model advantage is sample-size dependent and that must be reported.\n\nH5 (secondary — cross-atlas zero-shot generalisation): Foundation-model embeddings trained on Fullard alone achieve PD-vs-ctrl AUC ≥ 0.65 when scored zero-shot on Wang and Adams (no fine-tuning). Below this floor, the embedding is Fullard-specific rather than generalising.\n\nNull payoff. If neither Geneformer nor scGPT beats log-CPM-PCA-LR (ΔAUC < 0.05), the field's optimism about single-cell foundation models for neurodegenerative disease is overstated for PD — matching the Boiarsky 2024 + Kedzierska 2024 findings in other diseases, and placing PD on the benchmark map.", "report": null, "page": "/campaign/runs/parkinson-foundation-embedding-r2/" }, { "run": "parkinson-foundation-embedding-r3", "topic": "parkinson-foundation-embedding", "attempt": 3, "disease": "PD", "census_class": "PARTIAL", "final_status": "in_progress", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Do zero-shot Geneformer (95M-cell variant) and scGPT embeddings of Fullard 2024 SNpc dopaminergic-neuron nuclei separate PD from control donors better than a baseline log-CPM-PCA-LR pipeline at the donor level — and which foundation model wins?\n\nCui et al. 2024 (scGPT, Nat Methods 21:1470) and Theodoris et al. 2023+2024 (Geneformer, Nature 618:616 + 95M update) made strong claims about disease-stratification utility of foundation-model embeddings on cell atlases. Boiarsky 2024 and Kedzierska 2024 have begun to publish benchmarks showing zero-shot foundation models are often *worse* than baseline log-CPM-PCA-LR in disease contexts. PD has not been placed on that benchmark map. This OmiCairn run does so.\n\nHypotheses (pre-registered):\n\nH1 (primary — donor-level classification, foundation model vs. baseline): Donor-level leave-one-out (LOO) AUC for PD-vs-control classification using donor-mean foundation-model embedding + logistic regression is > 0.75 for at least one of Geneformer or scGPT, AND that AUC is > the log-CPM-PCA-LR baseline AUC by ΔAUC ≥ 0.05, with DeLong test p < 0.05.\n\nH2 (secondary — AGTR1+ subcluster recovery without supervision): Embedding-derived nearest-neighbour graph (Leiden γ = 0.6) in PD donors recovers an AGTR1+ vs. AGTR1- subcluster partition with Adjusted Rand Index ≥ 0.4 against the v1 `parkinson-da-vulnerability` workspace labels.\n\nH3 (secondary — transfer-learning benefit): Geneformer fine-tuned for 1 epoch on Kamath A1 PD-vs-ctrl labels (GSE178265) generalises to Wang (GSE184950) and Adams (GSE193688) atlases with held-out AUC ≥ 0.7 in each, vs. ≤ 0.6 for the same architecture trained from scratch on the same fine-tune set.\n\nH4 (secondary — embedding-space robustness): Repeat H1 with three subsampling levels (5K / 25K / 50K nuclei per donor average); the ΔAUC vs. baseline must be stable (within ±0.05) across the three subsamples, otherwise the foundation-model advantage is sample-size dependent and that must be reported.\n\nH5 (secondary — cross-atlas zero-shot generalisation): Foundation-model embeddings trained on Fullard alone achieve PD-vs-ctrl AUC ≥ 0.65 when scored zero-shot on Wang and Adams (no fine-tuning). Below this floor, the embedding is Fullard-specific rather than generalising.\n\nNull payoff. If neither Geneformer nor scGPT beats log-CPM-PCA-LR (ΔAUC < 0.05), the field's optimism about single-cell foundation models for neurodegenerative disease is overstated for PD — matching the Boiarsky 2024 + Kedzierska 2024 findings in other diseases, and placing PD on the benchmark map.", "report": null, "page": "/campaign/runs/parkinson-foundation-embedding-r3/" }, { "run": "parkinson-foundation-embedding-r4", "topic": "parkinson-foundation-embedding", "attempt": 4, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 0, "not_supported": 1, "inconclusive": 1, "not_testable": 2, "not_run": 1 }, "evaluated": 4, "stop_reason": null, "question": "Do zero-shot Geneformer (95M-cell variant) and scGPT embeddings of Fullard 2024 SNpc dopaminergic-neuron nuclei separate PD from control donors better than a baseline log-CPM-PCA-LR pipeline at the donor level — and which foundation model wins?\n\nCui et al. 2024 (scGPT, Nat Methods 21:1470) and Theodoris et al. 2023+2024 (Geneformer, Nature 618:616 + 95M update) made strong claims about disease-stratification utility of foundation-model embeddings on cell atlases. Boiarsky 2024 and Kedzierska 2024 have begun to publish benchmarks showing zero-shot foundation models are often *worse* than baseline log-CPM-PCA-LR in disease contexts. PD has not been placed on that benchmark map. This OmiCairn run does so.\n\nHypotheses (pre-registered):\n\nH1 (primary — donor-level classification, foundation model vs. baseline): Donor-level leave-one-out (LOO) AUC for PD-vs-control classification using donor-mean foundation-model embedding + logistic regression is > 0.75 for at least one of Geneformer or scGPT, AND that AUC is > the log-CPM-PCA-LR baseline AUC by ΔAUC ≥ 0.05, with DeLong test p < 0.05.\n\nH2 (secondary — AGTR1+ subcluster recovery without supervision): Embedding-derived nearest-neighbour graph (Leiden γ = 0.6) in PD donors recovers an AGTR1+ vs. AGTR1- subcluster partition with Adjusted Rand Index ≥ 0.4 against the v1 `parkinson-da-vulnerability` workspace labels.\n\nH3 (secondary — transfer-learning benefit): Geneformer fine-tuned for 1 epoch on Kamath A1 PD-vs-ctrl labels (GSE178265) generalises to Wang (GSE184950) and Adams (GSE193688) atlases with held-out AUC ≥ 0.7 in each, vs. ≤ 0.6 for the same architecture trained from scratch on the same fine-tune set.\n\nH4 (secondary — embedding-space robustness): Repeat H1 with three subsampling levels (5K / 25K / 50K nuclei per donor average); the ΔAUC vs. baseline must be stable (within ±0.05) across the three subsamples, otherwise the foundation-model advantage is sample-size dependent and that must be reported.\n\nH5 (secondary — cross-atlas zero-shot generalisation): Foundation-model embeddings trained on Fullard alone achieve PD-vs-ctrl AUC ≥ 0.65 when scored zero-shot on Wang and Adams (no fine-tuning). Below this floor, the embedding is Fullard-specific rather than generalising.\n\nNull payoff. If neither Geneformer nor scGPT beats log-CPM-PCA-LR (ΔAUC < 0.05), the field's optimism about single-cell foundation models for neurodegenerative disease is overstated for PD — matching the Boiarsky 2024 + Kedzierska 2024 findings in other diseases, and placing PD on the benchmark map.", "report": { "kind": "html", "url": "/campaign/runs/parkinson-foundation-embedding-r4/report/", "title": "Zero-shot foundation models do not improve donor-level Parkinson's disease classification (ΔAUC = +0.026)", "main_conclusion": "H1 not supported: Geneformer-V2-104M zero-shot LOO AUC 0.775 vs log-CPM-PCA-LR 0.749 (ΔAUC +0.026, 95% CI -0.066 to +0.118, p_dir 0.290, ~19% power, so the authors call it inconclusive rather than a confident null). Embeddings carry real signal (perm p<0.001). On Adams (n=13) the baseline dominates Geneformer (AUC 0.944 vs 0.694). The scGPT arm and H3 were deferred.", "thumb": "/assets/img/thumbs/parkinson-foundation-embedding-r4.webp" }, "page": "/campaign/runs/parkinson-foundation-embedding-r4/" }, { "run": "parkinson-gut-brain-axis-r1", "topic": "parkinson-gut-brain-axis", "attempt": 1, "disease": "PD", "census_class": "EARLY", "final_status": "halted", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Wallen, Z.D., Demirkan, A., Twa, G., ..., Payami, H. (Nat Commun 13:6958, 2022; DOI 10.1038/s41467-022-34667-x; SRA PRJNA834801; Zenodo 7246185) reported that >30% of bacterial species, genes, and metabolic pathways are altered in PD vs. controls in a shotgun-metagenomic cohort of 490 PD + 234 controls, describing a \"disease-permissive\" PD microbiome (elevated opportunistic pathogens, reduced SCFA producers, dysregulated neuroactive signaling).\n\nThe Fullard / N M et al. multi-region snRNA-seq atlas (Sci Data 11:1274, 2024; CELLxGENE collection d5d0df8f-4eee-49d8-a221- a288f50a1590; 100 donors, 5 regions, 2.1M nuclei) includes the dorsal motor nucleus of the vagus (DMNX) — the Braak stage 1 region that receives vagal afferents from the gut and is among the earliest CNS sites of Lewy pathology.\n\nDespite a decade of \"gut-brain-axis in PD\" review literature, no published study has formally tested whether the PD-gut-dysbiosis signature (Wallen 2022) is paralleled in DMNX-resident neuron transcriptomic dysregulation in PD (Fullard 2024 DMNX slab), and whether plasma DDC / immunometabolic proteomic markers (Bayati, Hällqvist) sit on the mechanistic path.\n\nBecause the Wallen and Fullard cohorts have no individual-donor overlap, this proposal does not test individual-level mediation but tests *signature-level* alignment via canonical correlation, pathway enrichment, and triple-omics meta-analysis. This is the informative design under public-data constraints.\n\nHypotheses (pre-registered):\n\nH1 (primary — bacterial-pathway-to-DMNX-gene-pathway enrichment alignment): The Wallen 2022 PD-vs-ctrl differential MetaCyc pathway list (q < 0.05 from ANCOM-BC, top 100 by effect size) and the Fullard DMNX-derived PD-vs-ctrl Reactome/KEGG enrichment list (computed on per-donor pseudobulk, top 100) share significantly more pathway-membership genes / shared KEGG modules than expected by chance, with hypergeometric p < 0.01. Concretely: of the top 100 Wallen pathways, ≥ 15 map via KEGG-Reactome bridging to KEGG modules that also appear in the top 100 Fullard DMNX enriched modules, where chance expectation is ≈ 6 under a hypergeometric null.\n\nH2 (secondary — vagal-target gene module alignment): A pre-defined \"vagal-afferent target\" gene module (vagal- sensory neuron markers + their transcriptional targets: TRPV1, ASIC3, P2RX2, CHRNA3, CHAT, PHOX2B, MNX1; their KEGG pathway \"Cholinergic synapse\" hsa04725 + \"Vagal innervation\" downstream targets) is differentially expressed in DMNX between PD and ctrl with FDR < 0.05, AND its expression direction is anti-correlated with the Wallen 2022 SCFA- producer abundance signature: PD = decreased SCFA producers + decreased vagal-target signal (Spearman ρ < -0.3 on random-pair permutation null over pseudo-cohort matched means).\n\nH3 (secondary — neuroinflammatory cross-modality): DMNX microglia in the Fullard atlas show PD-vs-ctrl up- regulation of the LPS-response and pattern-recognition- receptor (TLR4, TLR2, CD14, NLRP3) module, AND the magnitude of this upregulation correlates positively with the Wallen 2022 LPS-producer abundance signature when both are normalised to z-scores within their respective cohorts.\n\nH4 (secondary — protein-MR triangulation): Of the proteins nominated as causal for PD by the Storm 2025 Mendelian-randomisation analysis (UKB Olink × Nalls 2019 GWAS — public sumstats only, no individual-level UKB access required for the MR direction we test), the subset with bacterial-pathway-membership in Wallen 2022's PD- enriched pathways constitutes a \"triangulated\" panel; we test whether this panel size > expected by chance under permutation null over the full Storm-nominated protein set.\n\nH5 (secondary — alternative axes): We pre-register and test 3 *alternate* gene modules (catecholamine biosynthesis, ferroptosis pathway, cuproptosis pathway) as negative controls — these should NOT show signature alignment with gut-microbiome signatures via the same H1 methodology. If they DO, the H1 signal is suspect.\n\nA null finding on H1 would be informative — it would mean that signature-level gut-brain harmonisation is undetectable from bulk-pathway alignment alone, motivating per-individual multi-omic data acquisition (PPMI Project 196 + microbiome).", "report": null, "page": "/campaign/runs/parkinson-gut-brain-axis-r1/" }, { "run": "parkinson-gut-brain-axis-r2", "topic": "parkinson-gut-brain-axis", "attempt": 2, "disease": "PD", "census_class": "PARTIAL", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 1, "partial": 1, "not_supported": 3, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Wallen, Z.D., Demirkan, A., Twa, G., ..., Payami, H. (Nat Commun 13:6958, 2022; DOI 10.1038/s41467-022-34667-x; SRA PRJNA834801; Zenodo 7246185) reported that >30% of bacterial species, genes, and metabolic pathways are altered in PD vs. controls in a shotgun-metagenomic cohort of 490 PD + 234 controls, describing a \"disease-permissive\" PD microbiome (elevated opportunistic pathogens, reduced SCFA producers, dysregulated neuroactive signaling).\n\nThe Fullard / N M et al. multi-region snRNA-seq atlas (Sci Data 11:1274, 2024; CELLxGENE collection d5d0df8f-4eee-49d8-a221- a288f50a1590; 100 donors, 5 regions, 2.1M nuclei) includes the dorsal motor nucleus of the vagus (DMNX) — the Braak stage 1 region that receives vagal afferents from the gut and is among the earliest CNS sites of Lewy pathology.\n\nDespite a decade of \"gut-brain-axis in PD\" review literature, no published study has formally tested whether the PD-gut-dysbiosis signature (Wallen 2022) is paralleled in DMNX-resident neuron transcriptomic dysregulation in PD (Fullard 2024 DMNX slab), and whether plasma DDC / immunometabolic proteomic markers (Bayati, Hällqvist) sit on the mechanistic path.\n\nBecause the Wallen and Fullard cohorts have no individual-donor overlap, this proposal does not test individual-level mediation but tests *signature-level* alignment via canonical correlation, pathway enrichment, and triple-omics meta-analysis. This is the informative design under public-data constraints.\n\nHypotheses (pre-registered):\n\nH1 (primary — bacterial-pathway-to-DMNX-gene-pathway enrichment alignment): The Wallen 2022 PD-vs-ctrl differential MetaCyc pathway list (q < 0.05 from ANCOM-BC, top 100 by effect size) and the Fullard DMNX-derived PD-vs-ctrl Reactome/KEGG enrichment list (computed on per-donor pseudobulk, top 100) share significantly more pathway-membership genes / shared KEGG modules than expected by chance, with hypergeometric p < 0.01. Concretely: of the top 100 Wallen pathways, ≥ 15 map via KEGG-Reactome bridging to KEGG modules that also appear in the top 100 Fullard DMNX enriched modules, where chance expectation is ≈ 6 under a hypergeometric null.\n\nH2 (secondary — vagal-target gene module alignment): A pre-defined \"vagal-afferent target\" gene module (vagal- sensory neuron markers + their transcriptional targets: TRPV1, ASIC3, P2RX2, CHRNA3, CHAT, PHOX2B, MNX1; their KEGG pathway \"Cholinergic synapse\" hsa04725 + \"Vagal innervation\" downstream targets) is differentially expressed in DMNX between PD and ctrl with FDR < 0.05, AND its expression direction is anti-correlated with the Wallen 2022 SCFA- producer abundance signature: PD = decreased SCFA producers + decreased vagal-target signal (Spearman ρ < -0.3 on random-pair permutation null over pseudo-cohort matched means).\n\nH3 (secondary — neuroinflammatory cross-modality): DMNX microglia in the Fullard atlas show PD-vs-ctrl up- regulation of the LPS-response and pattern-recognition- receptor (TLR4, TLR2, CD14, NLRP3) module, AND the magnitude of this upregulation correlates positively with the Wallen 2022 LPS-producer abundance signature when both are normalised to z-scores within their respective cohorts.\n\nH4 (secondary — protein-MR triangulation): Of the proteins nominated as causal for PD by the Storm 2025 Mendelian-randomisation analysis (UKB Olink × Nalls 2019 GWAS — public sumstats only, no individual-level UKB access required for the MR direction we test), the subset with bacterial-pathway-membership in Wallen 2022's PD- enriched pathways constitutes a \"triangulated\" panel; we test whether this panel size > expected by chance under permutation null over the full Storm-nominated protein set.\n\nH5 (secondary — alternative axes): We pre-register and test 3 *alternate* gene modules (catecholamine biosynthesis, ferroptosis pathway, cuproptosis pathway) as negative controls — these should NOT show signature alignment with gut-microbiome signatures via the same H1 methodology. If they DO, the H1 signal is suspect.\n\nA null finding on H1 would be informative — it would mean that signature-level gut-brain harmonisation is undetectable from bulk-pathway alignment alone, motivating per-individual multi-omic data acquisition (PPMI Project 196 + microbiome).", "report": null, "page": "/campaign/runs/parkinson-gut-brain-axis-r2/" }, { "run": "parkinson-gut-brain-axis-r3", "topic": "parkinson-gut-brain-axis", "attempt": 3, "disease": "PD", "census_class": "EARLY", "final_status": "in_progress", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Wallen, Z.D., Demirkan, A., Twa, G., ..., Payami, H. (Nat Commun 13:6958, 2022; DOI 10.1038/s41467-022-34667-x; SRA PRJNA834801; Zenodo 7246185) reported that >30% of bacterial species, genes, and metabolic pathways are altered in PD vs. controls in a shotgun-metagenomic cohort of 490 PD + 234 controls, describing a \"disease-permissive\" PD microbiome (elevated opportunistic pathogens, reduced SCFA producers, dysregulated neuroactive signaling).\n\nThe Fullard / N M et al. multi-region snRNA-seq atlas (Sci Data 11:1274, 2024; CELLxGENE collection d5d0df8f-4eee-49d8-a221- a288f50a1590; 100 donors, 5 regions, 2.1M nuclei) includes the dorsal motor nucleus of the vagus (DMNX) — the Braak stage 1 region that receives vagal afferents from the gut and is among the earliest CNS sites of Lewy pathology.\n\nDespite a decade of \"gut-brain-axis in PD\" review literature, no published study has formally tested whether the PD-gut-dysbiosis signature (Wallen 2022) is paralleled in DMNX-resident neuron transcriptomic dysregulation in PD (Fullard 2024 DMNX slab), and whether plasma DDC / immunometabolic proteomic markers (Bayati, Hällqvist) sit on the mechanistic path.\n\nBecause the Wallen and Fullard cohorts have no individual-donor overlap, this proposal does not test individual-level mediation but tests *signature-level* alignment via canonical correlation, pathway enrichment, and triple-omics meta-analysis. This is the informative design under public-data constraints.\n\nHypotheses (pre-registered):\n\nH1 (primary — bacterial-pathway-to-DMNX-gene-pathway enrichment alignment): The Wallen 2022 PD-vs-ctrl differential MetaCyc pathway list (q < 0.05 from ANCOM-BC, top 100 by effect size) and the Fullard DMNX-derived PD-vs-ctrl Reactome/KEGG enrichment list (computed on per-donor pseudobulk, top 100) share significantly more pathway-membership genes / shared KEGG modules than expected by chance, with hypergeometric p < 0.01. Concretely: of the top 100 Wallen pathways, ≥ 15 map via KEGG-Reactome bridging to KEGG modules that also appear in the top 100 Fullard DMNX enriched modules, where chance expectation is ≈ 6 under a hypergeometric null.\n\nH2 (secondary — vagal-target gene module alignment): A pre-defined \"vagal-afferent target\" gene module (vagal- sensory neuron markers + their transcriptional targets: TRPV1, ASIC3, P2RX2, CHRNA3, CHAT, PHOX2B, MNX1; their KEGG pathway \"Cholinergic synapse\" hsa04725 + \"Vagal innervation\" downstream targets) is differentially expressed in DMNX between PD and ctrl with FDR < 0.05, AND its expression direction is anti-correlated with the Wallen 2022 SCFA- producer abundance signature: PD = decreased SCFA producers + decreased vagal-target signal (Spearman ρ < -0.3 on random-pair permutation null over pseudo-cohort matched means).\n\nH3 (secondary — neuroinflammatory cross-modality): DMNX microglia in the Fullard atlas show PD-vs-ctrl up- regulation of the LPS-response and pattern-recognition- receptor (TLR4, TLR2, CD14, NLRP3) module, AND the magnitude of this upregulation correlates positively with the Wallen 2022 LPS-producer abundance signature when both are normalised to z-scores within their respective cohorts.\n\nH4 (secondary — protein-MR triangulation): Of the proteins nominated as causal for PD by the Storm 2025 Mendelian-randomisation analysis (UKB Olink × Nalls 2019 GWAS — public sumstats only, no individual-level UKB access required for the MR direction we test), the subset with bacterial-pathway-membership in Wallen 2022's PD- enriched pathways constitutes a \"triangulated\" panel; we test whether this panel size > expected by chance under permutation null over the full Storm-nominated protein set.\n\nH5 (secondary — alternative axes): We pre-register and test 3 *alternate* gene modules (catecholamine biosynthesis, ferroptosis pathway, cuproptosis pathway) as negative controls — these should NOT show signature alignment with gut-microbiome signatures via the same H1 methodology. If they DO, the H1 signal is suspect.\n\nA null finding on H1 would be informative — it would mean that signature-level gut-brain harmonisation is undetectable from bulk-pathway alignment alone, motivating per-individual multi-omic data acquisition (PPMI Project 196 + microbiome).", "report": null, "page": "/campaign/runs/parkinson-gut-brain-axis-r3/" }, { "run": "parkinson-gut-brain-axis-r4", "topic": "parkinson-gut-brain-axis", "attempt": 4, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 1, "partial": 0, "not_supported": 4, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Wallen, Z.D., Demirkan, A., Twa, G., ..., Payami, H. (Nat Commun 13:6958, 2022; DOI 10.1038/s41467-022-34667-x; SRA PRJNA834801; Zenodo 7246185) reported that >30% of bacterial species, genes, and metabolic pathways are altered in PD vs. controls in a shotgun-metagenomic cohort of 490 PD + 234 controls, describing a \"disease-permissive\" PD microbiome (elevated opportunistic pathogens, reduced SCFA producers, dysregulated neuroactive signaling).\n\nThe Fullard / N M et al. multi-region snRNA-seq atlas (Sci Data 11:1274, 2024; CELLxGENE collection d5d0df8f-4eee-49d8-a221- a288f50a1590; 100 donors, 5 regions, 2.1M nuclei) includes the dorsal motor nucleus of the vagus (DMNX) — the Braak stage 1 region that receives vagal afferents from the gut and is among the earliest CNS sites of Lewy pathology.\n\nDespite a decade of \"gut-brain-axis in PD\" review literature, no published study has formally tested whether the PD-gut-dysbiosis signature (Wallen 2022) is paralleled in DMNX-resident neuron transcriptomic dysregulation in PD (Fullard 2024 DMNX slab), and whether plasma DDC / immunometabolic proteomic markers (Bayati, Hällqvist) sit on the mechanistic path.\n\nBecause the Wallen and Fullard cohorts have no individual-donor overlap, this proposal does not test individual-level mediation but tests *signature-level* alignment via canonical correlation, pathway enrichment, and triple-omics meta-analysis. This is the informative design under public-data constraints.\n\nHypotheses (pre-registered):\n\nH1 (primary — bacterial-pathway-to-DMNX-gene-pathway enrichment alignment): The Wallen 2022 PD-vs-ctrl differential MetaCyc pathway list (q < 0.05 from ANCOM-BC, top 100 by effect size) and the Fullard DMNX-derived PD-vs-ctrl Reactome/KEGG enrichment list (computed on per-donor pseudobulk, top 100) share significantly more pathway-membership genes / shared KEGG modules than expected by chance, with hypergeometric p < 0.01. Concretely: of the top 100 Wallen pathways, ≥ 15 map via KEGG-Reactome bridging to KEGG modules that also appear in the top 100 Fullard DMNX enriched modules, where chance expectation is ≈ 6 under a hypergeometric null.\n\nH2 (secondary — vagal-target gene module alignment): A pre-defined \"vagal-afferent target\" gene module (vagal- sensory neuron markers + their transcriptional targets: TRPV1, ASIC3, P2RX2, CHRNA3, CHAT, PHOX2B, MNX1; their KEGG pathway \"Cholinergic synapse\" hsa04725 + \"Vagal innervation\" downstream targets) is differentially expressed in DMNX between PD and ctrl with FDR < 0.05, AND its expression direction is anti-correlated with the Wallen 2022 SCFA- producer abundance signature: PD = decreased SCFA producers + decreased vagal-target signal (Spearman ρ < -0.3 on random-pair permutation null over pseudo-cohort matched means).\n\nH3 (secondary — neuroinflammatory cross-modality): DMNX microglia in the Fullard atlas show PD-vs-ctrl up- regulation of the LPS-response and pattern-recognition- receptor (TLR4, TLR2, CD14, NLRP3) module, AND the magnitude of this upregulation correlates positively with the Wallen 2022 LPS-producer abundance signature when both are normalised to z-scores within their respective cohorts.\n\nH4 (secondary — protein-MR triangulation): Of the proteins nominated as causal for PD by the Storm 2025 Mendelian-randomisation analysis (UKB Olink × Nalls 2019 GWAS — public sumstats only, no individual-level UKB access required for the MR direction we test), the subset with bacterial-pathway-membership in Wallen 2022's PD- enriched pathways constitutes a \"triangulated\" panel; we test whether this panel size > expected by chance under permutation null over the full Storm-nominated protein set.\n\nH5 (secondary — alternative axes): We pre-register and test 3 *alternate* gene modules (catecholamine biosynthesis, ferroptosis pathway, cuproptosis pathway) as negative controls — these should NOT show signature alignment with gut-microbiome signatures via the same H1 methodology. If they DO, the H1 signal is suspect.\n\nA null finding on H1 would be informative — it would mean that signature-level gut-brain harmonisation is undetectable from bulk-pathway alignment alone, motivating per-individual multi-omic data acquisition (PPMI Project 196 + microbiome).", "report": { "kind": "pdf", "url": "/assets/reports/parkinson-gut-brain-axis-r4.pdf", "bytes": 373174, "title": "Gut microbiome and DMNX transcriptomics share 10 of 131 KEGG modules in Parkinson's disease: a pre-registered null study", "main_conclusion": "Pre-registered null: 10 of 131 KEGG modules are shared between the Wallen gut and Fullard DMNX signatures (overlap 0.076, hypergeometric p=0.175, perm p=0.180), below the ≥15 threshold. H2, H3 and H4 were all non-significant (FDR 0.463, 0.514; p=0.373). Only the H5 negative controls passed (all p=1.0).", "thumb": "/assets/img/thumbs/parkinson-gut-brain-axis-r4.webp" }, "page": "/campaign/runs/parkinson-gut-brain-axis-r4/" }, { "run": "parkinson-lamda-multiregion-r1", "topic": "parkinson-lamda-multiregion", "attempt": 1, "disease": "PD", "census_class": "EARLY", "final_status": "halted", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Goralski, T.M., Henderson, M.X. et al. (Nat Commun 15:2642, 2024; DOI 10.1038/s41467-024-47027-8; PMC10966039) defined the \"Lewy- Associated Molecular Dysfunction from Aggregates\" (LAMDA) spatial- transcriptomic signature in PD/PDD/DLB *cingulate* cortex — 8 donors, NanoString GeoMx DSP whole-transcriptome + CosMx single- cell-resolved spatial profiling — and replicated the program in α-synuclein PFF-injected mouse cortex. They showed that pSyn-positive excitatory cortical neurons converge on a conserved dysfunction program (synaptic loss, mitochondrial dysfunction, proteostasis collapse, vesicle-trafficking failure).\n\nGoralski explicitly framed the result as cortically defined. It has *never* been tested whether LAMDA is detectable, in the same cell types or in different cell types, in *non-cortical* Lewy-pathology regions of Parkinson's brain — i.e., across the Braak progression from peripheral (DMNX), through midbrain (SNpc, GPi) to cortical (M1, V1) regions in the same donors.\n\nThe Fullard / N M et al. multi-region snRNA-seq atlas (Sci Data 11:1274, 2024; CELLxGENE collection d5d0df8f-4eee-49d8-a221- a288f50a1590) is the ideal test substrate: 100 donors (75 PD + 25 control), 5 regions in 71 of those donors (DMNX, GPi, SNpc, M1, V1), 2.1M nuclei total. Cingulate cortex is NOT part of the Fullard atlas — meaning the Goralski signature can be tested on fully independent tissue (positive-control cross-tissue spec test) and on regions with different Braak progression stages.\n\nHypotheses (pre-registered):\n\nH1 (primary — LAMDA generalises to non-cortical regions): Score the published Goralski LAMDA gene module (extracted from Goralski Supplementary Data 1 — top-200 upregulated genes in cortical pSyn+ neurons; Zenodo 10.5281/zenodo.10729767) as a per-cell module score across all 5 regions × all ~10 major cell types of the Fullard atlas. Per-(region × cell-type) PD-vs-control mean score difference is tested with linear mixed-effects model (donor as random intercept; sex + age fixed). After BH correction over the 50 tests, at least 3 NON-CORTICAL (region × cell-type) combinations show positive differential LAMDA enrichment at FDR < 0.05, with effect- size > 0.25 SD (Cohen's d).\n\nH2 (secondary — Braak-stage tracking): The LAMDA-enrichment gradient across regions in PD follows Braak progression: DMNX (Braak 1) > SNpc (Braak 3) > GPi (Braak 4) > M1 (Braak 5/6) > V1 (Braak 6), with monotonic ranking test (Spearman ρ on region-ordered mean LAMDA scores) significant at p < 0.05. (NB: this direction is the *opposite* of cortical-Lewy gradients; we test it explicitly because Goralski's LAMDA is derived from cortex, where it is presumed late. The null hypothesis is no monotonic gradient.)\n\nH3 (secondary — excitatory-neuron specificity replicates): Within each region's neuronal compartment, the LAMDA module score is significantly higher in excitatory neurons than inhibitory neurons (in cortical regions M1, V1), AND in glutamatergic projection neurons compared to local interneurons in subcortical regions (DMNX, SNpc, GPi). Wilcoxon test, BH-corrected per region.\n\nH4 (secondary — gene-program decomposition): Goralski LAMDA is a coarse 200-gene composite. We decompose into 5 functional sub-programs (synaptic / mitochondrial / proteostasis / vesicle / immune) per the Goralski-published functional clustering. We test whether the sub-programs dissociate across regions: do all 5 move together in cortex, but only some sub-programs activate in DMNX or SNpc? This dissociation pattern is the biological discovery target.\n\nH5 (secondary — direction-specific validation in cingulate held-out donor subset): Goralski has a subset of donors with both cingulate (their derivation tissue) and orbitofrontal data not used in the original LAMDA derivation. We hold out the orbitofrontal subset from Goralski's deposited CosMx data as a positive control: LAMDA should re-replicate at p < 0.001 in that tissue, providing a same-modality / same-cohort sanity bar.\n\nA non-generalisation of LAMDA outside cortex would itself be a major scientific finding — it would mean the program is cortex-restricted, contradicting common assumptions that Lewy-pathology gene programs are tissue-agnostic. OmiCairn must report any null finding with the same emphasis as a positive one.", "report": null, "page": "/campaign/runs/parkinson-lamda-multiregion-r1/" }, { "run": "parkinson-lamda-multiregion-r2", "topic": "parkinson-lamda-multiregion", "attempt": 2, "disease": "PD", "census_class": "EARLY", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 5 }, "evaluated": 0, "stop_reason": null, "question": "Goralski, T.M., Henderson, M.X. et al. (Nat Commun 15:2642, 2024; DOI 10.1038/s41467-024-47027-8; PMC10966039) defined the \"Lewy- Associated Molecular Dysfunction from Aggregates\" (LAMDA) spatial- transcriptomic signature in PD/PDD/DLB *cingulate* cortex — 8 donors, NanoString GeoMx DSP whole-transcriptome + CosMx single- cell-resolved spatial profiling — and replicated the program in α-synuclein PFF-injected mouse cortex. They showed that pSyn-positive excitatory cortical neurons converge on a conserved dysfunction program (synaptic loss, mitochondrial dysfunction, proteostasis collapse, vesicle-trafficking failure).\n\nGoralski explicitly framed the result as cortically defined. It has *never* been tested whether LAMDA is detectable, in the same cell types or in different cell types, in *non-cortical* Lewy-pathology regions of Parkinson's brain — i.e., across the Braak progression from peripheral (DMNX), through midbrain (SNpc, GPi) to cortical (M1, V1) regions in the same donors.\n\nThe Fullard / N M et al. multi-region snRNA-seq atlas (Sci Data 11:1274, 2024; CELLxGENE collection d5d0df8f-4eee-49d8-a221- a288f50a1590) is the ideal test substrate: 100 donors (75 PD + 25 control), 5 regions in 71 of those donors (DMNX, GPi, SNpc, M1, V1), 2.1M nuclei total. Cingulate cortex is NOT part of the Fullard atlas — meaning the Goralski signature can be tested on fully independent tissue (positive-control cross-tissue spec test) and on regions with different Braak progression stages.\n\nHypotheses (pre-registered):\n\nH1 (primary — LAMDA generalises to non-cortical regions): Score the published Goralski LAMDA gene module (extracted from Goralski Supplementary Data 1 — top-200 upregulated genes in cortical pSyn+ neurons; Zenodo 10.5281/zenodo.10729767) as a per-cell module score across all 5 regions × all ~10 major cell types of the Fullard atlas. Per-(region × cell-type) PD-vs-control mean score difference is tested with linear mixed-effects model (donor as random intercept; sex + age fixed). After BH correction over the 50 tests, at least 3 NON-CORTICAL (region × cell-type) combinations show positive differential LAMDA enrichment at FDR < 0.05, with effect- size > 0.25 SD (Cohen's d).\n\nH2 (secondary — Braak-stage tracking): The LAMDA-enrichment gradient across regions in PD follows Braak progression: DMNX (Braak 1) > SNpc (Braak 3) > GPi (Braak 4) > M1 (Braak 5/6) > V1 (Braak 6), with monotonic ranking test (Spearman ρ on region-ordered mean LAMDA scores) significant at p < 0.05. (NB: this direction is the *opposite* of cortical-Lewy gradients; we test it explicitly because Goralski's LAMDA is derived from cortex, where it is presumed late. The null hypothesis is no monotonic gradient.)\n\nH3 (secondary — excitatory-neuron specificity replicates): Within each region's neuronal compartment, the LAMDA module score is significantly higher in excitatory neurons than inhibitory neurons (in cortical regions M1, V1), AND in glutamatergic projection neurons compared to local interneurons in subcortical regions (DMNX, SNpc, GPi). Wilcoxon test, BH-corrected per region.\n\nH4 (secondary — gene-program decomposition): Goralski LAMDA is a coarse 200-gene composite. We decompose into 5 functional sub-programs (synaptic / mitochondrial / proteostasis / vesicle / immune) per the Goralski-published functional clustering. We test whether the sub-programs dissociate across regions: do all 5 move together in cortex, but only some sub-programs activate in DMNX or SNpc? This dissociation pattern is the biological discovery target.\n\nH5 (secondary — direction-specific validation in cingulate held-out donor subset): Goralski has a subset of donors with both cingulate (their derivation tissue) and orbitofrontal data not used in the original LAMDA derivation. We hold out the orbitofrontal subset from Goralski's deposited CosMx data as a positive control: LAMDA should re-replicate at p < 0.001 in that tissue, providing a same-modality / same-cohort sanity bar.\n\nA non-generalisation of LAMDA outside cortex would itself be a major scientific finding — it would mean the program is cortex-restricted, contradicting common assumptions that Lewy-pathology gene programs are tissue-agnostic. OmiCairn must report any null finding with the same emphasis as a positive one.", "report": null, "page": "/campaign/runs/parkinson-lamda-multiregion-r2/" }, { "run": "parkinson-lamda-multiregion-r3", "topic": "parkinson-lamda-multiregion", "attempt": 3, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 1, "not_supported": 3, "inconclusive": 0, "not_testable": 1, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Goralski, T.M., Henderson, M.X. et al. (Nat Commun 15:2642, 2024; DOI 10.1038/s41467-024-47027-8; PMC10966039) defined the \"Lewy- Associated Molecular Dysfunction from Aggregates\" (LAMDA) spatial- transcriptomic signature in PD/PDD/DLB *cingulate* cortex — 8 donors, NanoString GeoMx DSP whole-transcriptome + CosMx single- cell-resolved spatial profiling — and replicated the program in α-synuclein PFF-injected mouse cortex. They showed that pSyn-positive excitatory cortical neurons converge on a conserved dysfunction program (synaptic loss, mitochondrial dysfunction, proteostasis collapse, vesicle-trafficking failure).\n\nGoralski explicitly framed the result as cortically defined. It has *never* been tested whether LAMDA is detectable, in the same cell types or in different cell types, in *non-cortical* Lewy-pathology regions of Parkinson's brain — i.e., across the Braak progression from peripheral (DMNX), through midbrain (SNpc, GPi) to cortical (M1, V1) regions in the same donors.\n\nThe Fullard / N M et al. multi-region snRNA-seq atlas (Sci Data 11:1274, 2024; CELLxGENE collection d5d0df8f-4eee-49d8-a221- a288f50a1590) is the ideal test substrate: 100 donors (75 PD + 25 control), 5 regions in 71 of those donors (DMNX, GPi, SNpc, M1, V1), 2.1M nuclei total. Cingulate cortex is NOT part of the Fullard atlas — meaning the Goralski signature can be tested on fully independent tissue (positive-control cross-tissue spec test) and on regions with different Braak progression stages.\n\nHypotheses (pre-registered):\n\nH1 (primary — LAMDA generalises to non-cortical regions): Score the published Goralski LAMDA gene module (extracted from Goralski Supplementary Data 1 — top-200 upregulated genes in cortical pSyn+ neurons; Zenodo 10.5281/zenodo.10729767) as a per-cell module score across all 5 regions × all ~10 major cell types of the Fullard atlas. Per-(region × cell-type) PD-vs-control mean score difference is tested with linear mixed-effects model (donor as random intercept; sex + age fixed). After BH correction over the 50 tests, at least 3 NON-CORTICAL (region × cell-type) combinations show positive differential LAMDA enrichment at FDR < 0.05, with effect- size > 0.25 SD (Cohen's d).\n\nH2 (secondary — Braak-stage tracking): The LAMDA-enrichment gradient across regions in PD follows Braak progression: DMNX (Braak 1) > SNpc (Braak 3) > GPi (Braak 4) > M1 (Braak 5/6) > V1 (Braak 6), with monotonic ranking test (Spearman ρ on region-ordered mean LAMDA scores) significant at p < 0.05. (NB: this direction is the *opposite* of cortical-Lewy gradients; we test it explicitly because Goralski's LAMDA is derived from cortex, where it is presumed late. The null hypothesis is no monotonic gradient.)\n\nH3 (secondary — excitatory-neuron specificity replicates): Within each region's neuronal compartment, the LAMDA module score is significantly higher in excitatory neurons than inhibitory neurons (in cortical regions M1, V1), AND in glutamatergic projection neurons compared to local interneurons in subcortical regions (DMNX, SNpc, GPi). Wilcoxon test, BH-corrected per region.\n\nH4 (secondary — gene-program decomposition): Goralski LAMDA is a coarse 200-gene composite. We decompose into 5 functional sub-programs (synaptic / mitochondrial / proteostasis / vesicle / immune) per the Goralski-published functional clustering. We test whether the sub-programs dissociate across regions: do all 5 move together in cortex, but only some sub-programs activate in DMNX or SNpc? This dissociation pattern is the biological discovery target.\n\nH5 (secondary — direction-specific validation in cingulate held-out donor subset): Goralski has a subset of donors with both cingulate (their derivation tissue) and orbitofrontal data not used in the original LAMDA derivation. We hold out the orbitofrontal subset from Goralski's deposited CosMx data as a positive control: LAMDA should re-replicate at p < 0.001 in that tissue, providing a same-modality / same-cohort sanity bar.\n\nA non-generalisation of LAMDA outside cortex would itself be a major scientific finding — it would mean the program is cortex-restricted, contradicting common assumptions that Lewy-pathology gene programs are tissue-agnostic. OmiCairn must report any null finding with the same emphasis as a positive one.", "report": { "kind": "pdf", "url": "/assets/reports/parkinson-lamda-multiregion-r3.pdf", "bytes": 310388, "title": "Multi-region single-nucleus profiling in 97 Parkinson donors reveals cortex-restricted LAMDA gene program activation", "main_conclusion": "Null result: 0 of 22 non-cortical region x cell-type pairs showed positive LAMDA enrichment in PD (permutation p = 1.0; no better than shuffled-label or random-gene-set nulls). 49 of 53 effects were negative. H2 (Braak gradient), H4 (sub-program dissociation) and H5 (deferred) were not supported. The only positive result is EN > IN within cortex (d = 0.87-1.12, FDR < 3e-7).", "thumb": "/assets/img/thumbs/parkinson-lamda-multiregion-r3.webp" }, "page": "/campaign/runs/parkinson-lamda-multiregion-r3/" }, { "run": "parkinson-lincs-reversal-r1", "topic": "parkinson-lincs-reversal", "attempt": 1, "disease": "PD", "census_class": "PARTIAL", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 5 }, "evaluated": 0, "stop_reason": null, "question": "Which approved or late-phase compounds in the LINCS L1000 Phase II touchstone library produce gene-expression signatures that maximally reverse the PD-vs-control transcriptomic signature of SNpc dopaminergic neurons (Fullard 2024) — and how many of the top-50 reversing candidates have prior PD-drug-repositioning evidence in Open Targets or in published PD repurposing hit lists?\n\nLINCS L1000-based drug repositioning was pioneered by Iorio 2010 and applied broadly across diseases (CMAP score). PD applications have used bulk or whole-tissue signatures — never a cell-type-resolved, snRNA-seq-derived SNpc-DA-neuron signature from a multi-region atlas. The known concern is that LINCS L1000 cell lines are mostly cancer lines (MCF7, PC3, A375, etc.) and may be biologically irrelevant to neurons. This OmiCairn run tests exactly that.\n\nHypotheses (pre-registered):\n\nH1 (primary — top-50 reversers overlap with prior PD evidence): ≥ 5 of the top-50 negatively-correlated (most reversing) compounds by LINCS L1000 connectivity score against the SNpc-DA PD-vs-ctrl signature have prior Open Targets PD association (overall_association_score > 0 for MONDO:0005180) OR have appeared in a published PD drug-repositioning hit list (Bayati 2024 Acta Neuro DDC paper; omicSynth Tang 2024; Storm 2025). Hypergeometric q < 0.01 over background of all L1000 small-molecule perturbagens.\n\nH2 (secondary — drug-class enrichment): The top-50 hits are enriched for known PD-relevant therapeutic classes — anti-inflammatory, mTOR modulators, calcium-channel blockers, anti-oxidants, autophagy inducers. Pre-registered: ≥ 2 of these 5 classes over-represented by Fisher's exact q < 0.05.\n\nH3 (secondary — cell-type-specific reversal): Reversal scores in DA neurons differ from reversal scores in microglia from the same Fullard atlas (Pearson r < 0.3 for the two L1000 connectivity ranking lists). The optimal repurposing candidate is cell-type-dependent, not a global brain signal.\n\nH4 (secondary — neuronal-cell-line subset analysis): Restrict L1000 to its neuronal subset (NEU, NPC cell-line signatures only) — n = 2 cell lines, much smaller library. Test whether the neuronal-subset top-50 reversers have higher overlap with prior PD evidence (≥ 8 of 50) than the all-cell-lines top-50 (H1).\n\nH5 (secondary — sign-direction sanity check): Internal positive control: levodopa (DDC substrate) and ropinirole (D2 agonist) should appear in the *positive* connectivity tail (mimic PD-on-treatment state), NOT in the reversing tail. If they appear in the reversing tail, the signature direction is wrong.\n\nNull payoff. A null overlap (H1: 0/50 with prior PD evidence) means LINCS L1000 cancer-line transcriptomic signatures are too non-neuronal to be informative for PD — an informative quantification of the long-suspected limitation.", "report": null, "page": "/campaign/runs/parkinson-lincs-reversal-r1/" }, { "run": "parkinson-lincs-reversal-r2", "topic": "parkinson-lincs-reversal", "attempt": 2, "disease": "PD", "census_class": "PARTIAL", "final_status": "in_progress", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Which approved or late-phase compounds in the LINCS L1000 Phase II touchstone library produce gene-expression signatures that maximally reverse the PD-vs-control transcriptomic signature of SNpc dopaminergic neurons (Fullard 2024) — and how many of the top-50 reversing candidates have prior PD-drug-repositioning evidence in Open Targets or in published PD repurposing hit lists?\n\nLINCS L1000-based drug repositioning was pioneered by Iorio 2010 and applied broadly across diseases (CMAP score). PD applications have used bulk or whole-tissue signatures — never a cell-type-resolved, snRNA-seq-derived SNpc-DA-neuron signature from a multi-region atlas. The known concern is that LINCS L1000 cell lines are mostly cancer lines (MCF7, PC3, A375, etc.) and may be biologically irrelevant to neurons. This OmiCairn run tests exactly that.\n\nHypotheses (pre-registered):\n\nH1 (primary — top-50 reversers overlap with prior PD evidence): ≥ 5 of the top-50 negatively-correlated (most reversing) compounds by LINCS L1000 connectivity score against the SNpc-DA PD-vs-ctrl signature have prior Open Targets PD association (overall_association_score > 0 for MONDO:0005180) OR have appeared in a published PD drug-repositioning hit list (Bayati 2024 Acta Neuro DDC paper; omicSynth Tang 2024; Storm 2025). Hypergeometric q < 0.01 over background of all L1000 small-molecule perturbagens.\n\nH2 (secondary — drug-class enrichment): The top-50 hits are enriched for known PD-relevant therapeutic classes — anti-inflammatory, mTOR modulators, calcium-channel blockers, anti-oxidants, autophagy inducers. Pre-registered: ≥ 2 of these 5 classes over-represented by Fisher's exact q < 0.05.\n\nH3 (secondary — cell-type-specific reversal): Reversal scores in DA neurons differ from reversal scores in microglia from the same Fullard atlas (Pearson r < 0.3 for the two L1000 connectivity ranking lists). The optimal repurposing candidate is cell-type-dependent, not a global brain signal.\n\nH4 (secondary — neuronal-cell-line subset analysis): Restrict L1000 to its neuronal subset (NEU, NPC cell-line signatures only) — n = 2 cell lines, much smaller library. Test whether the neuronal-subset top-50 reversers have higher overlap with prior PD evidence (≥ 8 of 50) than the all-cell-lines top-50 (H1).\n\nH5 (secondary — sign-direction sanity check): Internal positive control: levodopa (DDC substrate) and ropinirole (D2 agonist) should appear in the *positive* connectivity tail (mimic PD-on-treatment state), NOT in the reversing tail. If they appear in the reversing tail, the signature direction is wrong.\n\nNull payoff. A null overlap (H1: 0/50 with prior PD evidence) means LINCS L1000 cancer-line transcriptomic signatures are too non-neuronal to be informative for PD — an informative quantification of the long-suspected limitation.", "report": null, "page": "/campaign/runs/parkinson-lincs-reversal-r2/" }, { "run": "parkinson-lincs-reversal-r3", "topic": "parkinson-lincs-reversal", "attempt": 3, "disease": "PD", "census_class": "PARTIAL", "final_status": "in_progress", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Which approved or late-phase compounds in the LINCS L1000 Phase II touchstone library produce gene-expression signatures that maximally reverse the PD-vs-control transcriptomic signature of SNpc dopaminergic neurons (Fullard 2024) — and how many of the top-50 reversing candidates have prior PD-drug-repositioning evidence in Open Targets or in published PD repurposing hit lists?\n\nLINCS L1000-based drug repositioning was pioneered by Iorio 2010 and applied broadly across diseases (CMAP score). PD applications have used bulk or whole-tissue signatures — never a cell-type-resolved, snRNA-seq-derived SNpc-DA-neuron signature from a multi-region atlas. The known concern is that LINCS L1000 cell lines are mostly cancer lines (MCF7, PC3, A375, etc.) and may be biologically irrelevant to neurons. This OmiCairn run tests exactly that.\n\nHypotheses (pre-registered):\n\nH1 (primary — top-50 reversers overlap with prior PD evidence): ≥ 5 of the top-50 negatively-correlated (most reversing) compounds by LINCS L1000 connectivity score against the SNpc-DA PD-vs-ctrl signature have prior Open Targets PD association (overall_association_score > 0 for MONDO:0005180) OR have appeared in a published PD drug-repositioning hit list (Bayati 2024 Acta Neuro DDC paper; omicSynth Tang 2024; Storm 2025). Hypergeometric q < 0.01 over background of all L1000 small-molecule perturbagens.\n\nH2 (secondary — drug-class enrichment): The top-50 hits are enriched for known PD-relevant therapeutic classes — anti-inflammatory, mTOR modulators, calcium-channel blockers, anti-oxidants, autophagy inducers. Pre-registered: ≥ 2 of these 5 classes over-represented by Fisher's exact q < 0.05.\n\nH3 (secondary — cell-type-specific reversal): Reversal scores in DA neurons differ from reversal scores in microglia from the same Fullard atlas (Pearson r < 0.3 for the two L1000 connectivity ranking lists). The optimal repurposing candidate is cell-type-dependent, not a global brain signal.\n\nH4 (secondary — neuronal-cell-line subset analysis): Restrict L1000 to its neuronal subset (NEU, NPC cell-line signatures only) — n = 2 cell lines, much smaller library. Test whether the neuronal-subset top-50 reversers have higher overlap with prior PD evidence (≥ 8 of 50) than the all-cell-lines top-50 (H1).\n\nH5 (secondary — sign-direction sanity check): Internal positive control: levodopa (DDC substrate) and ropinirole (D2 agonist) should appear in the *positive* connectivity tail (mimic PD-on-treatment state), NOT in the reversing tail. If they appear in the reversing tail, the signature direction is wrong.\n\nNull payoff. A null overlap (H1: 0/50 with prior PD evidence) means LINCS L1000 cancer-line transcriptomic signatures are too non-neuronal to be informative for PD — an informative quantification of the long-suspected limitation.", "report": null, "page": "/campaign/runs/parkinson-lincs-reversal-r3/" }, { "run": "parkinson-lincs-reversal-r4", "topic": "parkinson-lincs-reversal", "attempt": 4, "disease": "PD", "census_class": "SUCCESS", "final_status": "done", "n_hypotheses": 5, "counts": { "supported": 1, "partial": 0, "not_supported": 2, "inconclusive": 1, "not_testable": 1, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Which approved or late-phase compounds in the LINCS L1000 Phase II touchstone library produce gene-expression signatures that maximally reverse the PD-vs-control transcriptomic signature of SNpc dopaminergic neurons (Fullard 2024) — and how many of the top-50 reversing candidates have prior PD-drug-repositioning evidence in Open Targets or in published PD repurposing hit lists?\n\nLINCS L1000-based drug repositioning was pioneered by Iorio 2010 and applied broadly across diseases (CMAP score). PD applications have used bulk or whole-tissue signatures — never a cell-type-resolved, snRNA-seq-derived SNpc-DA-neuron signature from a multi-region atlas. The known concern is that LINCS L1000 cell lines are mostly cancer lines (MCF7, PC3, A375, etc.) and may be biologically irrelevant to neurons. This OmiCairn run tests exactly that.\n\nHypotheses (pre-registered):\n\nH1 (primary — top-50 reversers overlap with prior PD evidence): ≥ 5 of the top-50 negatively-correlated (most reversing) compounds by LINCS L1000 connectivity score against the SNpc-DA PD-vs-ctrl signature have prior Open Targets PD association (overall_association_score > 0 for MONDO:0005180) OR have appeared in a published PD drug-repositioning hit list (Bayati 2024 Acta Neuro DDC paper; omicSynth Tang 2024; Storm 2025). Hypergeometric q < 0.01 over background of all L1000 small-molecule perturbagens.\n\nH2 (secondary — drug-class enrichment): The top-50 hits are enriched for known PD-relevant therapeutic classes — anti-inflammatory, mTOR modulators, calcium-channel blockers, anti-oxidants, autophagy inducers. Pre-registered: ≥ 2 of these 5 classes over-represented by Fisher's exact q < 0.05.\n\nH3 (secondary — cell-type-specific reversal): Reversal scores in DA neurons differ from reversal scores in microglia from the same Fullard atlas (Pearson r < 0.3 for the two L1000 connectivity ranking lists). The optimal repurposing candidate is cell-type-dependent, not a global brain signal.\n\nH4 (secondary — neuronal-cell-line subset analysis): Restrict L1000 to its neuronal subset (NEU, NPC cell-line signatures only) — n = 2 cell lines, much smaller library. Test whether the neuronal-subset top-50 reversers have higher overlap with prior PD evidence (≥ 8 of 50) than the all-cell-lines top-50 (H1).\n\nH5 (secondary — sign-direction sanity check): Internal positive control: levodopa (DDC substrate) and ropinirole (D2 agonist) should appear in the *positive* connectivity tail (mimic PD-on-treatment state), NOT in the reversing tail. If they appear in the reversing tail, the signature direction is wrong.\n\nNull payoff. A null overlap (H1: 0/50 with prior PD evidence) means LINCS L1000 cancer-line transcriptomic signatures are too non-neuronal to be informative for PD — an informative quantification of the long-suspected limitation.", "report": { "kind": "pdf", "url": "/assets/reports/parkinson-lincs-reversal-r4.pdf", "bytes": 330453, "title": "Transcriptomic reversal of 2,429 compounds reveals dopaminergic--microglial divergence (r = 0.13) in Parkinson's disease", "main_conclusion": "The pre-registered primary endpoint (H1) was not met: 0 of 50 top reversers had prior PD evidence (q = 1.0). The report calls this an uninterpretable null because the evidence set was too small (M = 33 vs about 243 needed). H2 could not be assessed (CLUE API 401), H4 was not supported and the H5 positive control failed borderline (ropinirole at the 47.8th percentile). The only supported test was H3: DMNX_Neu vs microglia connectivity scores were weakly correlated (r = 0.133, 95% CI 0.094-0.172), which the report reads as cell-type-dependent drug prioritization.", "thumb": "/assets/img/thumbs/parkinson-lincs-reversal-r4.webp" }, "page": "/campaign/runs/parkinson-lincs-reversal-r4/" }, { "run": "parkinson-metabolic-flux-sn-r1", "topic": "parkinson-metabolic-flux-sn", "attempt": 1, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 7, "counts": { "supported": 0, "partial": 0, "not_supported": 1, "inconclusive": 5, "not_testable": 0, "not_run": 1 }, "evaluated": 6, "stop_reason": null, "question": "When the human genome-scale metabolic model Recon3D / Human-GEM is contextualised to PD-vs-control substantia-nigra pseudobulk expression (Fullard 2024 atlas, DA-neuron subset) via iMAT, do the predicted flux changes recapitulate the Supandi 2018 cortical-bulk findings (decreased glycolysis, decreased TCA flux midway, increased lactate) *specifically inside dopaminergic neurons* — a cell type that bulk microarray averages over?\n\nSupandi 2018 used PLOS One bulk microarray on PD cortex and predicted lower glycolytic and TCA flux. Their bulk preparation could not address SNpc DA neurons specifically. The 2024 Yamamori TIMBR paper (J Biotechnol Adv) extended GEM-based PD metabolic-biomarker prediction across 13 SN datasets, but remained bulk. Single-cell snRNA-seq atlases now make a DA-neuron-context constraint-based flux analysis possible. This OmiCairn run is the first to attempt it on Fullard data.\n\nHypotheses (pre-registered):\n\nH1 (primary — cortex-bulk concordance + DA glycolysis): In Recon3D contextualised by iMAT on Fullard SNpc DA pseudobulk, the PD-vs-control predicted flux change in cytoplasmic glycolysis (reactions PGK / HEX1 / PFK / ENO1 + 4 supporting reactions) is negative, with median signed flux ratio < 0.9 and bootstrap 95% CI excluding 1.0 (10,000 donor bootstraps). Sign concordance with Supandi 2018 cortical-bulk predictions on shared reactions is ≥ 70% (binomial test p < 0.01 against null 0.5).\n\nH2 (secondary — TIMBR biomarker recovery): TIMBR-predicted top-10 PD biomarker metabolites (Yamamori 2024) recover ≥ 4 metabolites whose abundance is also flagged by the Shao 2024 npj-PD plasma+CSF metabolomics meta-analysis (Supplementary Table 1) — a hypergeometric q < 0.05 cross-resource check.\n\nH3 (secondary — iron-metabolism elevation): Iron-metabolism flux (FECH-, FTH1-, TFRC-dependent reactions) is elevated in DA-neuron-context PD-vs-control flux ratio > 1.1 with bootstrap 95% CI excluding 1.0, consistent with the iron-accumulation vulnerability hypothesis (Hong 2025 PD-iron).\n\nH4 (secondary — contextualisation-method concordance): Repeat the analysis with three contextualisation methods — iMAT, TIMBR, E-Flux — and report Spearman ρ across per-reaction PD-vs-ctrl flux changes. ρ ≥ 0.5 between any two methods is the minimum agreement bar. (Per the v2 risk register, this is the dominant failure mode for constraint-based PD flux work.)\n\nH5 (secondary — DA-vs-glia cell-type separation): Repeat the analysis with Fullard microglia pseudobulk instead of DA-neuron pseudobulk. If H1's glycolysis-down finding is reproduced *only* in DA neurons and not in microglia, the bulk Supandi finding is DA-neuron-driven. If it is reproduced in both, bulk Supandi was averaging an organ-wide signal. Both outcomes are informative.\n\nNull payoff. A null DA-neuron H1 (median flux ratio ≥ 0.9 or bootstrap CI overlapping 1.0) means Supandi's bulk findings were driven by non-DA cell types (likely glia), not the vulnerable neurons themselves. Biologically informative.", "report": { "kind": "pdf", "url": "/assets/reports/parkinson-metabolic-flux-sn-r1.pdf", "bytes": 305891, "title": "Single-nucleus Recon3D Contextualization Reveals Direction-Consistent but Magnitude-Null Glycolytic Flux Changes in Parkinson's Disease Dopaminergic Neurons (83% Sign Concordance with Bulk, p=0.004; Cross-Method rho=0.94)", "main_conclusion": "Null primary result: the iMAT glycolysis flux ratio in DA neurons (PD/control) was 1.000 (95% CI 1.000-1.000; p = 1.0), so H1 was not supported. H1b (15/18 sign concordance, p = 0.0038) and H4 (iMAT vs E-Flux rho = 0.944) were confirmed only with caveats, as likely floating-point or model-topology artefacts. H2 and H3 were not supported and H5 was surrogate only. The report concludes the minimal 54-reaction ATPM model with n = 2 PD donors is structurally insufficient.", "thumb": "/assets/img/thumbs/parkinson-metabolic-flux-sn-r1.webp" }, "page": "/campaign/runs/parkinson-metabolic-flux-sn-r1/" }, { "run": "parkinson-multimodal-ppmi-trajectory-r1", "topic": "parkinson-multimodal-ppmi-trajectory", "attempt": 1, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 1, "not_supported": 1, "inconclusive": 2, "not_testable": 1, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Using only the publicly summarised longitudinal endpoints from PPMI — CSF α-syn-SAA Vmax/Tmax/AUC group means (Russo 2025 Lancet Neurology supplement), Olink Explore top-50-protein medians (Bayati 2024 Acta Neuropathologica supplement), motor MDS-UPDRS- III group-mean curves (Marek 2018 + downstream PPMI publications), free-water dMRI per-visit means (Burciu 2023 Frontiers Neurology supplement), LRRK2-G2019S SAA-stratified curves (Simuni 2024 Movement Disorders supplement) — can a Gaussian-process latent- variable model (GPLVM) infer a 1-D disease-time axis whose ordering recapitulates clinically known PD progression milestones?\n\nPPMI raw individual-level data are under a Data Use Agreement (DUA). Trajectory inference papers that re-derive from raw data cannot be reproduced without the DUA. This proposal asks the inverse methodological question: *how much disease-trajectory reconstruction is possible from supplementary-table summary statistics alone?* If the answer is \"a lot\", this is a reproducibility win and a template for PPMI-DUA-free research. If \"very little\", it puts a quantitative upper bound on summary-table-only PD work and justifies the DUA application.\n\nHypotheses (pre-registered):\n\nH1 (primary — disease-time correlates with time-from-diagnosis): The GPLVM-inferred 1-D latent disease-time axis correlates with published per-visit time-from-diagnosis at the cohort-summary level with Spearman ρ > 0.6 (n ≥ 50 visit-aggregated points across baseline / Y1 / Y2 / Y3 / Y5 timepoints × multiple cohort-subgroup strata).\n\nH2 (secondary — modality leave-one-out generalisation): Free-water dMRI (Burciu 2023) is held out; GPLVM trained on SAA + Olink + MDS-UPDRS only recovers free-water dMRI rank-order with ρ ≥ 0.4. (Tests that disease-time is modality-general, not modality-specific.)\n\nH3 (secondary — LRRK2 leftward-shift): LRRK2-G2019S carriers (Simuni 2024) have a leftward- shifted inferred disease-time distribution vs. sporadic PD at matched clinical-stage (Cohen's d > 0.5, two-sample Welch t-test p < 0.05). Consistent with the SAA-negative LRRK2 subtype being molecularly distinct.\n\nH4 (secondary — modality-specific weight check): Report per-modality GPLVM loading magnitude on the primary latent factor. No single modality should contribute > 70% of the loading. If one does, the \"multi-modal\" claim is misleading and that modality alone would do.\n\nH5 (secondary — supplementary-table fetchability audit): Of the 6 cited supplements (Russo, Bayati, Hällqvist, Burciu, Simuni, Marek), how many are (a) machine- readable (TSV / XLSX), (b) PDF-table only, (c) image only / not fetchable? This number itself is an informative methodological finding about PD trajectory work reproducibility.\n\nNull payoff. If summary tables don't encode enough information to reconstruct disease time (H1 ρ < 0.4), the PPMI raw-data DUA application is justified — and the proposal would put a hard number on what summary-table-only PD trajectory work can deliver.", "report": { "kind": "pdf", "url": "/assets/reports/parkinson-multimodal-ppmi-trajectory-r1.pdf", "bytes": 298737, "title": "PPMI supplement tables lack per-visit longitudinal data needed for disease-trajectory inference: a pre-registered audit", "main_conclusion": "Negative/inconclusive: public PPMI supplements do not encode enough information for disease-trajectory reconstruction. H1 is INCONCLUSIVE (n=10 against the pre-registered n>=50; FactorAnalysis rho=0.430, 95% CI [-0.459, 0.953], p_perm=0.118). None of the 6 supplements has per-visit longitudinal means (3 XLSX/TSV, 2 PDF-table, 1 figure-only). Proteomics makes up about 100% of the factor loading, so H4 is violated. H2 and H3 could not be evaluated.", "thumb": "/assets/img/thumbs/parkinson-multimodal-ppmi-trajectory-r1.webp" }, "page": "/campaign/runs/parkinson-multimodal-ppmi-trajectory-r1/" }, { "run": "parkinson-multimodal-spatial-proteomics-r1", "topic": "parkinson-multimodal-spatial-proteomics", "attempt": 1, "disease": "PD", "census_class": "PARTIAL", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 5 }, "evaluated": 0, "stop_reason": null, "question": "Does a MOFA+ (Multi-Omics-Factor-Analysis) joint embedding of three modalities of PD substantia-nigra data — Martirosyan 2024 snRNA-seq + Visium (GSE243639), Ma 2025 GeoMx spatial (PMC12322057), and Zhu 2024 LC-MS/MS DLPFC proteomics (PRIDE PXD030142) — reveal factor(s) that ONLY appear when all three modalities are jointly modelled, and that are unrecoverable from any pairwise integration?\n\nMulti-modal integration has become a default expectation in PD research, but the value-add of going from 2-modal to 3-modal is rarely tested with a clean ablation. Pairwise integrations in PD exist: snRNA × Visium (Martirosyan 2024); snRNA × proteomics (Zhu 2024); spatial × CosMx (Goralski 2024). Joint 3-modal MOFA+ on PD-SN has not been published. This OmiCairn run pre-registers the ablation that asks whether the third modality is necessary or decorative.\n\nHypotheses (pre-registered):\n\nH1 (primary — all-modalities factor exists): ≥ 1 MOFA+ factor explains > 5% variance in EACH of the three modalities simultaneously, AND is associated with PD diagnosis (per-donor-group Wilcoxon p < 0.05). Pre- registered name: \"all-modalities factor\".\n\nH2 (secondary — proteomics necessity ablation): Removing the proteomics modality and re-fitting MOFA+ produces a closest-matching factor (max absolute loading correlation with the H1 factor) whose explained variance in the spatial modality drops by ≥ 30 percentage points. Proteomics is NECESSARY for the joint factor, not nice-to-have.\n\nH3 (secondary — top-loaded gene enrichment): Top-100 loaded genes of the all-modalities factor are enriched (q < 0.05 hypergeometric) for the union of: Goralski 2024 LAMDA cortical signature genes, AND Hemedan 2024 Boolean-PD \"key-regulator\" gene list.\n\nH4 (secondary — modality-balance check): No single modality contributes > 60% of the explained variance summed across the H1 factor (otherwise the \"all-modalities\" label is misleading).\n\nH5 (secondary — pairwise comparison): Repeat the analysis with three pairwise MOFA+ models — (snRNA + Visium), (snRNA + proteomics), (Visium + proteomics). Test whether any pairwise model recovers a factor with ≥ 0.7 absolute loading correlation to the H1 3-modal factor. If yes, the 3-modal claim is weaker than headline. If no, 3-modal is necessary.\n\nNull payoff. A null all-modalities factor (H1: no factor >5% variance in all three modalities, or no PD association) means PD SN biology is dominated by the strongest single modality (probably snRNA-seq) and multi-omics doesn't add. Canonical ablation answer the field has been asking.", "report": null, "page": "/campaign/runs/parkinson-multimodal-spatial-proteomics-r1/" }, { "run": "parkinson-multimodal-spatial-proteomics-r2", "topic": "parkinson-multimodal-spatial-proteomics", "attempt": 2, "disease": "PD", "census_class": "PARTIAL", "final_status": "in_progress", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Does a MOFA+ (Multi-Omics-Factor-Analysis) joint embedding of three modalities of PD substantia-nigra data — Martirosyan 2024 snRNA-seq + Visium (GSE243639), Ma 2025 GeoMx spatial (PMC12322057), and Zhu 2024 LC-MS/MS DLPFC proteomics (PRIDE PXD030142) — reveal factor(s) that ONLY appear when all three modalities are jointly modelled, and that are unrecoverable from any pairwise integration?\n\nMulti-modal integration has become a default expectation in PD research, but the value-add of going from 2-modal to 3-modal is rarely tested with a clean ablation. Pairwise integrations in PD exist: snRNA × Visium (Martirosyan 2024); snRNA × proteomics (Zhu 2024); spatial × CosMx (Goralski 2024). Joint 3-modal MOFA+ on PD-SN has not been published. This OmiCairn run pre-registers the ablation that asks whether the third modality is necessary or decorative.\n\nHypotheses (pre-registered):\n\nH1 (primary — all-modalities factor exists): ≥ 1 MOFA+ factor explains > 5% variance in EACH of the three modalities simultaneously, AND is associated with PD diagnosis (per-donor-group Wilcoxon p < 0.05). Pre- registered name: \"all-modalities factor\".\n\nH2 (secondary — proteomics necessity ablation): Removing the proteomics modality and re-fitting MOFA+ produces a closest-matching factor (max absolute loading correlation with the H1 factor) whose explained variance in the spatial modality drops by ≥ 30 percentage points. Proteomics is NECESSARY for the joint factor, not nice-to-have.\n\nH3 (secondary — top-loaded gene enrichment): Top-100 loaded genes of the all-modalities factor are enriched (q < 0.05 hypergeometric) for the union of: Goralski 2024 LAMDA cortical signature genes, AND Hemedan 2024 Boolean-PD \"key-regulator\" gene list.\n\nH4 (secondary — modality-balance check): No single modality contributes > 60% of the explained variance summed across the H1 factor (otherwise the \"all-modalities\" label is misleading).\n\nH5 (secondary — pairwise comparison): Repeat the analysis with three pairwise MOFA+ models — (snRNA + Visium), (snRNA + proteomics), (Visium + proteomics). Test whether any pairwise model recovers a factor with ≥ 0.7 absolute loading correlation to the H1 3-modal factor. If yes, the 3-modal claim is weaker than headline. If no, 3-modal is necessary.\n\nNull payoff. A null all-modalities factor (H1: no factor >5% variance in all three modalities, or no PD association) means PD SN biology is dominated by the strongest single modality (probably snRNA-seq) and multi-omics doesn't add. Canonical ablation answer the field has been asking.", "report": null, "page": "/campaign/runs/parkinson-multimodal-spatial-proteomics-r2/" }, { "run": "parkinson-multimodal-spatial-proteomics-r3", "topic": "parkinson-multimodal-spatial-proteomics", "attempt": 3, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 5, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Does a MOFA+ (Multi-Omics-Factor-Analysis) joint embedding of three modalities of PD substantia-nigra data — Martirosyan 2024 snRNA-seq + Visium (GSE243639), Ma 2025 GeoMx spatial (PMC12322057), and Zhu 2024 LC-MS/MS DLPFC proteomics (PRIDE PXD030142) — reveal factor(s) that ONLY appear when all three modalities are jointly modelled, and that are unrecoverable from any pairwise integration?\n\nMulti-modal integration has become a default expectation in PD research, but the value-add of going from 2-modal to 3-modal is rarely tested with a clean ablation. Pairwise integrations in PD exist: snRNA × Visium (Martirosyan 2024); snRNA × proteomics (Zhu 2024); spatial × CosMx (Goralski 2024). Joint 3-modal MOFA+ on PD-SN has not been published. This OmiCairn run pre-registers the ablation that asks whether the third modality is necessary or decorative.\n\nHypotheses (pre-registered):\n\nH1 (primary — all-modalities factor exists): ≥ 1 MOFA+ factor explains > 5% variance in EACH of the three modalities simultaneously, AND is associated with PD diagnosis (per-donor-group Wilcoxon p < 0.05). Pre- registered name: \"all-modalities factor\".\n\nH2 (secondary — proteomics necessity ablation): Removing the proteomics modality and re-fitting MOFA+ produces a closest-matching factor (max absolute loading correlation with the H1 factor) whose explained variance in the spatial modality drops by ≥ 30 percentage points. Proteomics is NECESSARY for the joint factor, not nice-to-have.\n\nH3 (secondary — top-loaded gene enrichment): Top-100 loaded genes of the all-modalities factor are enriched (q < 0.05 hypergeometric) for the union of: Goralski 2024 LAMDA cortical signature genes, AND Hemedan 2024 Boolean-PD \"key-regulator\" gene list.\n\nH4 (secondary — modality-balance check): No single modality contributes > 60% of the explained variance summed across the H1 factor (otherwise the \"all-modalities\" label is misleading).\n\nH5 (secondary — pairwise comparison): Repeat the analysis with three pairwise MOFA+ models — (snRNA + Visium), (snRNA + proteomics), (Visium + proteomics). Test whether any pairwise model recovers a factor with ≥ 0.7 absolute loading correlation to the H1 3-modal factor. If yes, the 3-modal claim is weaker than headline. If no, 3-modal is necessary.\n\nNull payoff. A null all-modalities factor (H1: no factor >5% variance in all three modalities, or no PD association) means PD SN biology is dominated by the strongest single modality (probably snRNA-seq) and multi-omics doesn't add. Canonical ablation answer the field has been asking.", "report": { "kind": "pdf", "url": "/assets/reports/parkinson-multimodal-spatial-proteomics-r3.pdf", "bytes": 503770, "title": "3MAP: a pre-registered three-modal MOFA+ audit protocol for Parkinson's disease substantia nigra reveals data-access barriers and an 11.7-fold LAMDA-enriched dopaminergic factor", "main_conclusion": "Spatial and proteomics per-donor matrices could not be accessed, so only an snRNA-seq surrogate was run. Factor 1 explains 26.5% of variance but has no PD association (Wilcoxon p=0.948; H1 not supported). It is 11.7-fold enriched for LAMDA∪Hemedan PD genes (q_BH=5.07e-9; H3 supported). H2, H4 and H5 could not be evaluated.", "thumb": "/assets/img/thumbs/parkinson-multimodal-spatial-proteomics-r3.webp" }, "page": "/campaign/runs/parkinson-multimodal-spatial-proteomics-r3/" }, { "run": "parkinson-plasma-prodromal-panel-r1", "topic": "parkinson-plasma-prodromal-panel", "attempt": 1, "disease": "PD", "census_class": "EARLY", "final_status": "halted", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Hällqvist J., Bartl M., Dakna M., ..., Mollenhauer B., Trenkwalder C. (Nat Commun 15:4759, 2024; DOI 10.1038/s41467-024- 48961-3; PMC11189460; PRIDE PXD046355) reported that an 8-protein plasma panel (Granulin/GRN, MASP2, HSPA5/BiP, PTGDS, ICAM1, complement C3, DKK3, SERPING1) distinguishes individuals with isolated REM-sleep behaviour disorder (iRBD) who will convert to Parkinson's disease within 7 years from those who will not, with area under the curve up to 1.0 in their internal cohort.\n\nThe result is rare in PD biomarker literature: a small, interpretable panel measured by LC-MS/MS DIA in real plasma from a longitudinal prodromal cohort (DeNoPa + Oxford Discovery). It has not been formally replicated on the deposited MS spectra by an independent analysis pipeline, and the cross-platform generalisability (LC-MS/MS → other quantification) has not been established with deposited public data only.\n\nThis OmiCairn run formally tests:\n\nH1 (primary — re-analysis of the deposited MS spectra replicates the 8-protein panel's predictive performance): Re-processing PXD046355 spectra with a from-scratch DIA-NN / Spectronaut-equivalent pipeline (DIA-NN 1.9.2 or Skyline) → protein-level intensities → log-transform + quantile normalisation → 5-fold cross-validated logistic regression with the 8 published proteins as features, labels = iRBD-converter vs. iRBD-non-converter at 7-year follow-up. Primary success = 5-fold CV AUC ≥ 0.85 (the published Hällqvist 7-year AUC is ≈ 1.0; we allow margin for spectra → intensity pipeline divergence).\n\nH2 (secondary — individual-protein hazard ratios reproduce): Each of the 8 proteins, modeled univariately in a Cox proportional-hazards model on time-to-conversion, shows the same sign of hazard ratio as published (Hällqvist Supplementary Table 3). Concordance metric = 8/8 sign agreement; loose acceptance = ≥ 6/8.\n\nH3 (secondary — panel beats clinical-only baseline): Adding the 8-protein panel to a clinical-baseline Cox model (age + sex + MoCA + UPSIT) improves time-dependent AUC at 5 years by ΔAUC > 0.05 with paired bootstrap CI excluding 0. The clinical baseline serves as the Venet-style floor.\n\nH4 (secondary — out-of-cohort signature alignment to UK Biobank Olink): Of the 8 panel proteins, 6 are on the UK Biobank Olink Explore 1536 panel. We test whether the directionality (HR sign) reported for these 6 proteins in the Gan/You 2025 UKB Olink incident-PD analysis (Nat Aging 5:642) matches the Hällqvist signs. Concordance = 6/6 expected; loose acceptance ≥ 4/6 with binomial p < 0.05 vs. chance 0.5.\n\nH5 (secondary — sub-panel decomposition): Decompose the 8-panel into 3 functional sub-panels: (a) inflammation / complement: ICAM1, C3, SERPING1 (b) proteostasis: HSPA5, GRN (c) lipid / cell signaling: PTGDS, DKK3, MASP2 Test whether any single sub-panel achieves AUC > 0.75 on its own — i.e., is the predictive signal carried by one of three functional axes or distributed?\n\nA negative-leaning result on H1 (cross-pipeline AUC < 0.7) would be a major methodological caveat for the field and would imply that quantitative biomarker panels are sensitive to MS pipeline choice — itself an informative finding given the rapid uptake of the Hällqvist panel in trial design.", "report": null, "page": "/campaign/runs/parkinson-plasma-prodromal-panel-r1/" }, { "run": "parkinson-plasma-prodromal-panel-r2", "topic": "parkinson-plasma-prodromal-panel", "attempt": 2, "disease": "PD", "census_class": "PARTIAL", "final_status": "in_progress", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Hällqvist J., Bartl M., Dakna M., ..., Mollenhauer B., Trenkwalder C. (Nat Commun 15:4759, 2024; DOI 10.1038/s41467-024- 48961-3; PMC11189460; PRIDE PXD046355) reported that an 8-protein plasma panel (Granulin/GRN, MASP2, HSPA5/BiP, PTGDS, ICAM1, complement C3, DKK3, SERPING1) distinguishes individuals with isolated REM-sleep behaviour disorder (iRBD) who will convert to Parkinson's disease within 7 years from those who will not, with area under the curve up to 1.0 in their internal cohort.\n\nThe result is rare in PD biomarker literature: a small, interpretable panel measured by LC-MS/MS DIA in real plasma from a longitudinal prodromal cohort (DeNoPa + Oxford Discovery). It has not been formally replicated on the deposited MS spectra by an independent analysis pipeline, and the cross-platform generalisability (LC-MS/MS → other quantification) has not been established with deposited public data only.\n\nThis OmiCairn run formally tests:\n\nH1 (primary — re-analysis of the deposited MS spectra replicates the 8-protein panel's predictive performance): Re-processing PXD046355 spectra with a from-scratch DIA-NN / Spectronaut-equivalent pipeline (DIA-NN 1.9.2 or Skyline) → protein-level intensities → log-transform + quantile normalisation → 5-fold cross-validated logistic regression with the 8 published proteins as features, labels = iRBD-converter vs. iRBD-non-converter at 7-year follow-up. Primary success = 5-fold CV AUC ≥ 0.85 (the published Hällqvist 7-year AUC is ≈ 1.0; we allow margin for spectra → intensity pipeline divergence).\n\nH2 (secondary — individual-protein hazard ratios reproduce): Each of the 8 proteins, modeled univariately in a Cox proportional-hazards model on time-to-conversion, shows the same sign of hazard ratio as published (Hällqvist Supplementary Table 3). Concordance metric = 8/8 sign agreement; loose acceptance = ≥ 6/8.\n\nH3 (secondary — panel beats clinical-only baseline): Adding the 8-protein panel to a clinical-baseline Cox model (age + sex + MoCA + UPSIT) improves time-dependent AUC at 5 years by ΔAUC > 0.05 with paired bootstrap CI excluding 0. The clinical baseline serves as the Venet-style floor.\n\nH4 (secondary — out-of-cohort signature alignment to UK Biobank Olink): Of the 8 panel proteins, 6 are on the UK Biobank Olink Explore 1536 panel. We test whether the directionality (HR sign) reported for these 6 proteins in the Gan/You 2025 UKB Olink incident-PD analysis (Nat Aging 5:642) matches the Hällqvist signs. Concordance = 6/6 expected; loose acceptance ≥ 4/6 with binomial p < 0.05 vs. chance 0.5.\n\nH5 (secondary — sub-panel decomposition): Decompose the 8-panel into 3 functional sub-panels: (a) inflammation / complement: ICAM1, C3, SERPING1 (b) proteostasis: HSPA5, GRN (c) lipid / cell signaling: PTGDS, DKK3, MASP2 Test whether any single sub-panel achieves AUC > 0.75 on its own — i.e., is the predictive signal carried by one of three functional axes or distributed?\n\nA negative-leaning result on H1 (cross-pipeline AUC < 0.7) would be a major methodological caveat for the field and would imply that quantitative biomarker panels are sensitive to MS pipeline choice — itself an informative finding given the rapid uptake of the Hällqvist panel in trial design.", "report": null, "page": "/campaign/runs/parkinson-plasma-prodromal-panel-r2/" }, { "run": "parkinson-plasma-prodromal-panel-r3", "topic": "parkinson-plasma-prodromal-panel", "attempt": 3, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 0, "not_supported": 1, "inconclusive": 0, "not_testable": 4, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Hällqvist J., Bartl M., Dakna M., ..., Mollenhauer B., Trenkwalder C. (Nat Commun 15:4759, 2024; DOI 10.1038/s41467-024- 48961-3; PMC11189460; PRIDE PXD046355) reported that an 8-protein plasma panel (Granulin/GRN, MASP2, HSPA5/BiP, PTGDS, ICAM1, complement C3, DKK3, SERPING1) distinguishes individuals with isolated REM-sleep behaviour disorder (iRBD) who will convert to Parkinson's disease within 7 years from those who will not, with area under the curve up to 1.0 in their internal cohort.\n\nThe result is rare in PD biomarker literature: a small, interpretable panel measured by LC-MS/MS DIA in real plasma from a longitudinal prodromal cohort (DeNoPa + Oxford Discovery). It has not been formally replicated on the deposited MS spectra by an independent analysis pipeline, and the cross-platform generalisability (LC-MS/MS → other quantification) has not been established with deposited public data only.\n\nThis OmiCairn run formally tests:\n\nH1 (primary — re-analysis of the deposited MS spectra replicates the 8-protein panel's predictive performance): Re-processing PXD046355 spectra with a from-scratch DIA-NN / Spectronaut-equivalent pipeline (DIA-NN 1.9.2 or Skyline) → protein-level intensities → log-transform + quantile normalisation → 5-fold cross-validated logistic regression with the 8 published proteins as features, labels = iRBD-converter vs. iRBD-non-converter at 7-year follow-up. Primary success = 5-fold CV AUC ≥ 0.85 (the published Hällqvist 7-year AUC is ≈ 1.0; we allow margin for spectra → intensity pipeline divergence).\n\nH2 (secondary — individual-protein hazard ratios reproduce): Each of the 8 proteins, modeled univariately in a Cox proportional-hazards model on time-to-conversion, shows the same sign of hazard ratio as published (Hällqvist Supplementary Table 3). Concordance metric = 8/8 sign agreement; loose acceptance = ≥ 6/8.\n\nH3 (secondary — panel beats clinical-only baseline): Adding the 8-protein panel to a clinical-baseline Cox model (age + sex + MoCA + UPSIT) improves time-dependent AUC at 5 years by ΔAUC > 0.05 with paired bootstrap CI excluding 0. The clinical baseline serves as the Venet-style floor.\n\nH4 (secondary — out-of-cohort signature alignment to UK Biobank Olink): Of the 8 panel proteins, 6 are on the UK Biobank Olink Explore 1536 panel. We test whether the directionality (HR sign) reported for these 6 proteins in the Gan/You 2025 UKB Olink incident-PD analysis (Nat Aging 5:642) matches the Hällqvist signs. Concordance = 6/6 expected; loose acceptance ≥ 4/6 with binomial p < 0.05 vs. chance 0.5.\n\nH5 (secondary — sub-panel decomposition): Decompose the 8-panel into 3 functional sub-panels: (a) inflammation / complement: ICAM1, C3, SERPING1 (b) proteostasis: HSPA5, GRN (c) lipid / cell signaling: PTGDS, DKK3, MASP2 Test whether any single sub-panel achieves AUC > 0.75 on its own — i.e., is the predictive signal carried by one of three functional axes or distributed?\n\nA negative-leaning result on H1 (cross-pipeline AUC < 0.7) would be a major methodological caveat for the field and would imply that quantitative biomarker panels are sensitive to MS pipeline choice — itself an informative finding given the rapid uptake of the Hällqvist panel in trial design.", "report": { "kind": "html", "url": "/campaign/runs/parkinson-plasma-prodromal-panel-r3/report/", "title": "Independent replication of a plasma prodromal Parkinson panel: infrastructure barriers and cross-platform concordance", "main_conclusion": "Primary validation of the Hällqvist 8-protein panel was blocked: the PRIDE API returned empty data and DIA-NN 1.9.2 is Linux-only. For H4, 3/6 protein directions were concordant with UKB Olink incident-PD HRs (binomial p=0.656, CI 0.118-0.882), below the >=4/6 threshold, so H4 is not supported.", "thumb": "/assets/img/thumbs/parkinson-plasma-prodromal-panel-r3.webp" }, "page": "/campaign/runs/parkinson-plasma-prodromal-panel-r3/" }, { "run": "parkinson-plasma-prodromal-panel-r4", "topic": "parkinson-plasma-prodromal-panel", "attempt": 4, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 0, "not_supported": 2, "inconclusive": 1, "not_testable": 2, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Hällqvist J., Bartl M., Dakna M., ..., Mollenhauer B., Trenkwalder C. (Nat Commun 15:4759, 2024; DOI 10.1038/s41467-024- 48961-3; PMC11189460; PRIDE PXD046355) reported that an 8-protein plasma panel (Granulin/GRN, MASP2, HSPA5/BiP, PTGDS, ICAM1, complement C3, DKK3, SERPING1) distinguishes individuals with isolated REM-sleep behaviour disorder (iRBD) who will convert to Parkinson's disease within 7 years from those who will not, with area under the curve up to 1.0 in their internal cohort.\n\nThe result is rare in PD biomarker literature: a small, interpretable panel measured by LC-MS/MS DIA in real plasma from a longitudinal prodromal cohort (DeNoPa + Oxford Discovery). It has not been formally replicated on the deposited MS spectra by an independent analysis pipeline, and the cross-platform generalisability (LC-MS/MS → other quantification) has not been established with deposited public data only.\n\nThis OmiCairn run formally tests:\n\nH1 (primary — re-analysis of the deposited MS spectra replicates the 8-protein panel's predictive performance): Re-processing PXD046355 spectra with a from-scratch DIA-NN / Spectronaut-equivalent pipeline (DIA-NN 1.9.2 or Skyline) → protein-level intensities → log-transform + quantile normalisation → 5-fold cross-validated logistic regression with the 8 published proteins as features, labels = iRBD-converter vs. iRBD-non-converter at 7-year follow-up. Primary success = 5-fold CV AUC ≥ 0.85 (the published Hällqvist 7-year AUC is ≈ 1.0; we allow margin for spectra → intensity pipeline divergence).\n\nH2 (secondary — individual-protein hazard ratios reproduce): Each of the 8 proteins, modeled univariately in a Cox proportional-hazards model on time-to-conversion, shows the same sign of hazard ratio as published (Hällqvist Supplementary Table 3). Concordance metric = 8/8 sign agreement; loose acceptance = ≥ 6/8.\n\nH3 (secondary — panel beats clinical-only baseline): Adding the 8-protein panel to a clinical-baseline Cox model (age + sex + MoCA + UPSIT) improves time-dependent AUC at 5 years by ΔAUC > 0.05 with paired bootstrap CI excluding 0. The clinical baseline serves as the Venet-style floor.\n\nH4 (secondary — out-of-cohort signature alignment to UK Biobank Olink): Of the 8 panel proteins, 6 are on the UK Biobank Olink Explore 1536 panel. We test whether the directionality (HR sign) reported for these 6 proteins in the Gan/You 2025 UKB Olink incident-PD analysis (Nat Aging 5:642) matches the Hällqvist signs. Concordance = 6/6 expected; loose acceptance ≥ 4/6 with binomial p < 0.05 vs. chance 0.5.\n\nH5 (secondary — sub-panel decomposition): Decompose the 8-panel into 3 functional sub-panels: (a) inflammation / complement: ICAM1, C3, SERPING1 (b) proteostasis: HSPA5, GRN (c) lipid / cell signaling: PTGDS, DKK3, MASP2 Test whether any single sub-panel achieves AUC > 0.75 on its own — i.e., is the predictive signal carried by one of three functional axes or distributed?\n\nA negative-leaning result on H1 (cross-pipeline AUC < 0.7) would be a major methodological caveat for the field and would imply that quantitative biomarker panels are sensitive to MS pipeline choice — itself an informative finding given the rapid uptake of the Hällqvist panel in trial design.", "report": { "kind": "pdf", "url": "/assets/reports/parkinson-plasma-prodromal-panel-r4.pdf", "bytes": 376362, "title": "Label contamination, not panel failure, explains AUC 0.55 replication of an 8-protein prodromal Parkinson plasma panel", "main_conclusion": "The pre-registered 8-protein panel failed to replicate on PXD046355 MOESM4 (5-fold CV AUC = 0.554, 95% CI [0.437, 0.678], p = 0.254 vs permuted null; H1 not supported). The report attributes this to label contamination, based on a post-hoc clean-negative analysis (14 never-converters, AUC = 0.925). H2, H4 and H5 were not supported and H3 was blocked.", "thumb": "/assets/img/thumbs/parkinson-plasma-prodromal-panel-r4.webp" }, "page": "/campaign/runs/parkinson-plasma-prodromal-panel-r4/" }, { "run": "parkinson-prs-portability-r1", "topic": "parkinson-prs-portability", "attempt": 1, "disease": "PD", "census_class": "PARTIAL", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 5 }, "evaluated": 0, "stop_reason": null, "question": "When a polygenic risk score (PRS) is built from the Kim 2024 multi-ancestry PD GWAS (GCST90275127) via LDpred-funct on a 1000-Genomes LD reference, does its expected liability-scale R² in 1000 Genomes individuals stratified by Saffie-Awad-2025- style super-population ancestry-clusters (EUR, EAS, AMR, AFR, AJ) match the Saffie-Awad-2025-reported AUC values within ±0.05 per ancestry — purely using public sumstats + 1000 Genomes individual genotypes (no UK Biobank, no GP2 individual access)?\n\nSaffie-Awad 2025 trained 105 PRSs across 7 ancestries on GP2 R9 individual data and reported AUC by ancestry (EUR 0.621, EAS 0.578, AAC 0.591, AJ 0.67, AFR 0.583, AMR 0.639, CAS 0.612). Reproducing those AUCs requires GP2 access. This proposal asks the methodological question: how close can a *summary-statistics- only + 1000-Genomes-only* re-implementation get? The answer is itself an informative methodological contribution about what portability benchmarks are possible without IRB-gated cohort access.\n\nHypotheses (pre-registered):\n\nH1 (primary — EUR portability matches Saffie-Awad): PRS liability-scale R² in 1000 Genomes EUR individuals (CEU+GBR+FIN+IBS+TSI, n ≈ 503) is within ±0.05 of the liability-R² corresponding to Saffie-Awad 2025's Model 2 EUR AUC of 0.621 (translated via Lee 2012 formula given EUR PD prevalence 0.3%).\n\nH2 (secondary — ancestry-decay ratio): Ratio of AUC(EAS) / AUC(EUR) is within ±0.1 of the Saffie-Awad-reported 0.578 / 0.621 = 0.93. (Ancestry- decay must reproduce roughly, not exactly.)\n\nH3 (secondary — protein-MR PRS overlay): Adding the Storm 2025 protein-MR pQTL-instrumented gene- set as a fixed-effect overlay (PROteinMR-PRS) increases EUR liability-scale R² by ΔR² ≥ 0.01 vs. SNP-only PRS. Methodological test of whether protein-MR adds incremental information.\n\nH4 (secondary — multi-ancestry GWAS does not hurt EUR): The Nalls 2019 EUR-only PRS (GCST009325) does NOT significantly outperform the Kim 2024 multi-ancestry PRS in the EUR 1000 Genomes stratum (DeLong p > 0.1). Tests the field concern that multi-ancestry GWAS hurts EUR-stratum performance.\n\nH5 (secondary — bootstrap CI honesty): Per-ancestry bootstrap 95% CI is reported alongside point estimates. For ancestries with n < 200 in 1000 Genomes (e.g., AMR sub-populations), CI width > 0.15 is the expected outcome; this must be reported in `analysis summary` and `negative_results.md`, not buried.\n\nNull payoff. If LD-mismatch and small 1000 Genomes per-ancestry sample sizes inflate error bars beyond ±0.05 (H1), the Wang & Visscher 2024 concern about summary-only PRS portability benchmarking is confirmed, and the run would put a quantitative number on it. Either outcome is informative.", "report": null, "page": "/campaign/runs/parkinson-prs-portability-r1/" }, { "run": "parkinson-prs-portability-r2", "topic": "parkinson-prs-portability", "attempt": 2, "disease": "PD", "census_class": "PARTIAL", "final_status": "in_progress", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "When a polygenic risk score (PRS) is built from the Kim 2024 multi-ancestry PD GWAS (GCST90275127) via LDpred-funct on a 1000-Genomes LD reference, does its expected liability-scale R² in 1000 Genomes individuals stratified by Saffie-Awad-2025- style super-population ancestry-clusters (EUR, EAS, AMR, AFR, AJ) match the Saffie-Awad-2025-reported AUC values within ±0.05 per ancestry — purely using public sumstats + 1000 Genomes individual genotypes (no UK Biobank, no GP2 individual access)?\n\nSaffie-Awad 2025 trained 105 PRSs across 7 ancestries on GP2 R9 individual data and reported AUC by ancestry (EUR 0.621, EAS 0.578, AAC 0.591, AJ 0.67, AFR 0.583, AMR 0.639, CAS 0.612). Reproducing those AUCs requires GP2 access. This proposal asks the methodological question: how close can a *summary-statistics- only + 1000-Genomes-only* re-implementation get? The answer is itself an informative methodological contribution about what portability benchmarks are possible without IRB-gated cohort access.\n\nHypotheses (pre-registered):\n\nH1 (primary — EUR portability matches Saffie-Awad): PRS liability-scale R² in 1000 Genomes EUR individuals (CEU+GBR+FIN+IBS+TSI, n ≈ 503) is within ±0.05 of the liability-R² corresponding to Saffie-Awad 2025's Model 2 EUR AUC of 0.621 (translated via Lee 2012 formula given EUR PD prevalence 0.3%).\n\nH2 (secondary — ancestry-decay ratio): Ratio of AUC(EAS) / AUC(EUR) is within ±0.1 of the Saffie-Awad-reported 0.578 / 0.621 = 0.93. (Ancestry- decay must reproduce roughly, not exactly.)\n\nH3 (secondary — protein-MR PRS overlay): Adding the Storm 2025 protein-MR pQTL-instrumented gene- set as a fixed-effect overlay (PROteinMR-PRS) increases EUR liability-scale R² by ΔR² ≥ 0.01 vs. SNP-only PRS. Methodological test of whether protein-MR adds incremental information.\n\nH4 (secondary — multi-ancestry GWAS does not hurt EUR): The Nalls 2019 EUR-only PRS (GCST009325) does NOT significantly outperform the Kim 2024 multi-ancestry PRS in the EUR 1000 Genomes stratum (DeLong p > 0.1). Tests the field concern that multi-ancestry GWAS hurts EUR-stratum performance.\n\nH5 (secondary — bootstrap CI honesty): Per-ancestry bootstrap 95% CI is reported alongside point estimates. For ancestries with n < 200 in 1000 Genomes (e.g., AMR sub-populations), CI width > 0.15 is the expected outcome; this must be reported in `analysis summary` and `negative_results.md`, not buried.\n\nNull payoff. If LD-mismatch and small 1000 Genomes per-ancestry sample sizes inflate error bars beyond ±0.05 (H1), the Wang & Visscher 2024 concern about summary-only PRS portability benchmarking is confirmed, and the run would put a quantitative number on it. Either outcome is informative.", "report": null, "page": "/campaign/runs/parkinson-prs-portability-r2/" }, { "run": "parkinson-prs-portability-r3", "topic": "parkinson-prs-portability", "attempt": 3, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 1, "partial": 0, "not_supported": 2, "inconclusive": 1, "not_testable": 1, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "When a polygenic risk score (PRS) is built from the Kim 2024 multi-ancestry PD GWAS (GCST90275127) via LDpred-funct on a 1000-Genomes LD reference, does its expected liability-scale R² in 1000 Genomes individuals stratified by Saffie-Awad-2025- style super-population ancestry-clusters (EUR, EAS, AMR, AFR, AJ) match the Saffie-Awad-2025-reported AUC values within ±0.05 per ancestry — purely using public sumstats + 1000 Genomes individual genotypes (no UK Biobank, no GP2 individual access)?\n\nSaffie-Awad 2025 trained 105 PRSs across 7 ancestries on GP2 R9 individual data and reported AUC by ancestry (EUR 0.621, EAS 0.578, AAC 0.591, AJ 0.67, AFR 0.583, AMR 0.639, CAS 0.612). Reproducing those AUCs requires GP2 access. This proposal asks the methodological question: how close can a *summary-statistics- only + 1000-Genomes-only* re-implementation get? The answer is itself an informative methodological contribution about what portability benchmarks are possible without IRB-gated cohort access.\n\nHypotheses (pre-registered):\n\nH1 (primary — EUR portability matches Saffie-Awad): PRS liability-scale R² in 1000 Genomes EUR individuals (CEU+GBR+FIN+IBS+TSI, n ≈ 503) is within ±0.05 of the liability-R² corresponding to Saffie-Awad 2025's Model 2 EUR AUC of 0.621 (translated via Lee 2012 formula given EUR PD prevalence 0.3%).\n\nH2 (secondary — ancestry-decay ratio): Ratio of AUC(EAS) / AUC(EUR) is within ±0.1 of the Saffie-Awad-reported 0.578 / 0.621 = 0.93. (Ancestry- decay must reproduce roughly, not exactly.)\n\nH3 (secondary — protein-MR PRS overlay): Adding the Storm 2025 protein-MR pQTL-instrumented gene- set as a fixed-effect overlay (PROteinMR-PRS) increases EUR liability-scale R² by ΔR² ≥ 0.01 vs. SNP-only PRS. Methodological test of whether protein-MR adds incremental information.\n\nH4 (secondary — multi-ancestry GWAS does not hurt EUR): The Nalls 2019 EUR-only PRS (GCST009325) does NOT significantly outperform the Kim 2024 multi-ancestry PRS in the EUR 1000 Genomes stratum (DeLong p > 0.1). Tests the field concern that multi-ancestry GWAS hurts EUR-stratum performance.\n\nH5 (secondary — bootstrap CI honesty): Per-ancestry bootstrap 95% CI is reported alongside point estimates. For ancestries with n < 200 in 1000 Genomes (e.g., AMR sub-populations), CI width > 0.15 is the expected outcome; this must be reported in `analysis summary` and `negative_results.md`, not buried.\n\nNull payoff. If LD-mismatch and small 1000 Genomes per-ancestry sample sizes inflate error bars beyond ±0.05 (H1), the Wang & Visscher 2024 concern about summary-only PRS portability benchmarking is confirmed, and the run would put a quantitative number on it. Either outcome is informative.", "report": { "kind": "pdf", "url": "/assets/reports/parkinson-prs-portability-r3.pdf", "bytes": 315714, "title": "OPENVAL-PD: An Open Benchmarking Protocol for Polygenic Risk Score Portability in Parkinson's Disease Using Public Sumstats and 1000 Genomes Allele Frequencies", "main_conclusion": "A summary-statistics-only LDpred-inf PRS (Kim 2024) gives an analytical EUR AUC of 0.548, 0.073 below the Saffie-Awad benchmark of 0.621 (H1 rejected). H2 is supported only because of a genome-build mismatch artefact. H4 is reversed (Kim beats Nalls, confounded by an N_eff metadata artefact), H3 is deferred and H5 is rejected.", "thumb": "/assets/img/thumbs/parkinson-prs-portability-r3.webp" }, "page": "/campaign/runs/parkinson-prs-portability-r3/" }, { "run": "parkinson-scenic-regulons-r1", "topic": "parkinson-scenic-regulons", "attempt": 1, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 1, "partial": 0, "not_supported": 0, "inconclusive": 2, "not_testable": 2, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Which transcription-factor (TF) regulons, inferred de novo by pySCENIC from the Fullard 2024 multi-region snRNA-seq atlas (CELLxGENE d5d0df8f-...) on substantia-nigra pars compacta (SNpc) dopaminergic (DA) neurons, are differentially active between Parkinson's-disease (PD) and age-matched control donors — and do those regulon-activity shifts replicate in the independent Smajic / Wang / Adams SNpc atlases when the same regulon dictionary is scored zero-shot via AUCell?\n\nSelective-vulnerability biology in PD has so far been read at the cluster + marker-gene resolution (Kamath 2022 AGTR1+; Wang 2024 RIT2+/CALB1-). The TF-network layer above the markers has not been formally tested for *cross-atlas reproducibility* — i.e., is the dysregulation a coherent regulon shift, or only an apparent marker-gene shift that scrambles when the network is recomputed from a different atlas? This OmiCairn run pre-registers that test.\n\nHypotheses (pre-registered):\n\nH1 (primary — regulon replication across three independent atlases): Let R = the top-20 SNpc-DA PD-vs-control differentially active regulons in the Fullard atlas (FDR < 0.05, limma- voom on the donor × regulon-AUCell matrix with age + sex + post-mortem interval covariates). At least 3 of those 20 regulons recover a same-sign log-fold-change in regulon- AUCell score in *all three* held-out atlases (Smajic GSE157783, Wang GSE184950, Adams GSE193688), with a binomial test p < 0.01 against a null of 0.5 sign-agreement for each regulon × atlas combination.\n\nH2 (secondary — TF gene-level GWAS enrichment): The PD-up-regulated regulon set (FDR < 0.05 in Fullard, same-sign in ≥ 2 of 3 replication atlases) is enriched by hypergeometric test (q < 0.05) for TFs whose genes carry PD-risk fine-mapped variants in the Kim 2024 multi-ancestry GWAS (GCST90275127 Supplementary Table S5).\n\nH3 (secondary — alignment with AGTR1+ depletion): Per-donor AUCell scores of the consensus PD-up regulons rank-correlate (Spearman ρ > 0.3, p < 0.05) with the AGTR1+ DA depletion score derived in the v1 `parkinson-da-vulnerability` workspace (read from prior workspace `results record`), at the donor level inside Fullard SNpc.\n\nH4 (secondary — donor-level classification): Leave-one-atlas-out donor-level random forest using only the top-50 SNpc DA regulon AUCell scores as features achieves an AUC ≥ 0.70 for PD-vs-control classification (95% bootstrap CI excluding 0.5).\n\nH5 (secondary — clustering-resolution robustness): Re-run the top-level Leiden clustering at γ ∈ {0.4, 0.8, 1.2}; the H1 same-sign replication count varies by ≤ 1 regulon across the three resolutions (the dominant failure mode the v1 parkinson-da-vulnerability case logged).\n\nNull payoff. If consensus regulons fail to replicate across atlases (H1 = 0/20 same-sign in ≥ 2 atlases), the field's TF-level mechanistic claim is at the cluster-resolution layer, not the regulon-network layer. That is an informative Tier-A negative.", "report": { "kind": "pdf", "url": "/assets/reports/parkinson-scenic-regulons-r1.pdf", "bytes": 294013, "title": "Pre-registered TF regulon benchmarking across Parkinson's disease SNpc atlases yields a null replication signal", "main_conclusion": "Top-20 TFs from the Adams atlas (4 PD donors) did not replicate in the Wang atlas: 9/20 had the same sign (binomial p = 0.748; Δ vs shuffled regulons = -0.928, permutation p = 0.731). H1 was falsified on the single available atlas, H2 and H3 were inapplicable, and H4 was not supported. H5 was supported: the null is stable across Leiden resolutions.", "thumb": "/assets/img/thumbs/parkinson-scenic-regulons-r1.webp" }, "page": "/campaign/runs/parkinson-scenic-regulons-r1/" }, { "run": "parkinson-tcell-foothold-r1", "topic": "parkinson-tcell-foothold", "attempt": 1, "disease": "PD", "census_class": "EARLY", "final_status": "halted", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": "Ma, J., Al-Dalahmah, O., Goldman, J.E., Sulzer, D., Sims, P.A. et al. (Nat Commun s41467-025-62478-3, 2025; PMC12322057) reported spatially resolved adaptive immunity in Parkinson's disease substantia nigra: combined snRNA-seq + 10x Visium + scTCR-seq across 44 donors (13 PD + 15 ctrl SN; 10 PD + 8 ctrl cingulate; 265,284 nuclei + 10 donors with Visium) identified clonally expanded CD8+ T cells whose TCRs share homology with α-synuclein peptides and which spatially co-localise with CD44+ reactive astrocytes in PD SN.\n\nThe Ma 2025 result depends on rare TCR-seq data, which the Kamath 2022 (GSE178265), Wang 2024 (GSE184950), Adams 2024 (GSE193688) and Fullard 2024 multi-region atlases do not have. The question is whether the transcriptomic *footprint* of the Ma-2025 T-cell signature (the genes co-expressed in the clonally expanded CD8+ population) can be detected in those four independent atlases, in a deconvolution-inferred T-cell or lymphocyte cluster, without TCR ground truth.\n\nHypotheses (pre-registered):\n\nH1 (primary — transcriptomic footprint replicates in independent PD snRNA-seq atlases): The Ma 2025-defined T-cell signature (extracted from Supplementary Data of PMC12322057 — the top-50 differentially expressed genes in the clonally expanded CD8+ population vs. all other immune nuclei in PD SN) is scored per cell in T-cell / lymphocyte clusters of:\n\n- Kamath 2022 (GSE178265) SNpc + caudate\n\n- Wang 2024 (GSE184950) SNpc\n\n- Adams 2024 (GSE193688) SN\n\n- Fullard 2024 SNpc + M1 (CELLxGENE d5d0df8f-...) and tested for PD-vs-control per-donor pseudobulk differential at FDR<0.1. At least 3 of 4 atlases show log2FC > 0 in the T-cell-module score in the PD direction.\n\nH2 (secondary — region specificity within multi-region atlas): Within the Fullard atlas (the only multi-region atlas in the panel), the T-cell-module signal is significantly higher in SNpc + cingulate-like cortical regions than in DMNX (Braak 1, vagal-driven, no expected adaptive immune component) or V1 (Braak 6, late-stage, no expected ongoing immune infiltration). Pre-registered ranking: SNpc > M1 > V1 ≈ DMNX. Spearman ρ on rank vs. signature score > 0.5, permutation p < 0.05.\n\nH3 (secondary — T-cell-module × DA-loss covariation): At the donor level, in Kamath / Wang / Adams / Fullard atlases, the T-cell-module pseudobulk score is positively correlated with the AGTR1+ DA-loss score (computed identically to parkinson-da-vulnerability proposal): donors with more T-cell signal also have more vulnerable- DA loss. Spearman ρ > 0.3, permutation p < 0.05 in ≥ 2 of 4 atlases.\n\nH4 (secondary — astrocyte co-localisation in spatial slabs): Where Visium-spatial data is available (Martirosyan A5, Ma 2025 itself), spatial spots scoring high for the T-cell module are *also* enriched for CD44+ / GFAP-high reactive astrocyte markers (Ma 2025's reported co-localisation), with Spearman ρ > 0.3 on per-spot z-scored module scores.\n\nH5 (secondary — non-α-syn-homology bound — negative control): We pre-register a non-PD-relevant T-cell sub-signature (a generic exhaustion module: PDCD1, HAVCR2, LAG3, TOX, TIGIT) as a negative control. The exhaustion module should NOT show preferential PD enrichment (this signal is canonical to tumor immunology and PD T cells are not exhausted in the same way). A positive enrichment of the exhaustion module would suggest the H1 signal is a non-specific \"T cell present\" signal rather than a PD-specific clonally-expanded program.\n\nA null result on H1 is informative: it would mean that without TCR-seq ground truth, the clonal-expansion signal in Ma 2025 is not detectable from gene expression alone — calling into question any retrospective inference of adaptive immunity from existing PD snRNA-seq atlases.", "report": null, "page": "/campaign/runs/parkinson-tcell-foothold-r1/" }, { "run": "parkinson-tcell-foothold-r2", "topic": "parkinson-tcell-foothold", "attempt": 2, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 3, "not_testable": 2, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Ma, J., Al-Dalahmah, O., Goldman, J.E., Sulzer, D., Sims, P.A. et al. (Nat Commun s41467-025-62478-3, 2025; PMC12322057) reported spatially resolved adaptive immunity in Parkinson's disease substantia nigra: combined snRNA-seq + 10x Visium + scTCR-seq across 44 donors (13 PD + 15 ctrl SN; 10 PD + 8 ctrl cingulate; 265,284 nuclei + 10 donors with Visium) identified clonally expanded CD8+ T cells whose TCRs share homology with α-synuclein peptides and which spatially co-localise with CD44+ reactive astrocytes in PD SN.\n\nThe Ma 2025 result depends on rare TCR-seq data, which the Kamath 2022 (GSE178265), Wang 2024 (GSE184950), Adams 2024 (GSE193688) and Fullard 2024 multi-region atlases do not have. The question is whether the transcriptomic *footprint* of the Ma-2025 T-cell signature (the genes co-expressed in the clonally expanded CD8+ population) can be detected in those four independent atlases, in a deconvolution-inferred T-cell or lymphocyte cluster, without TCR ground truth.\n\nHypotheses (pre-registered):\n\nH1 (primary — transcriptomic footprint replicates in independent PD snRNA-seq atlases): The Ma 2025-defined T-cell signature (extracted from Supplementary Data of PMC12322057 — the top-50 differentially expressed genes in the clonally expanded CD8+ population vs. all other immune nuclei in PD SN) is scored per cell in T-cell / lymphocyte clusters of:\n\n- Kamath 2022 (GSE178265) SNpc + caudate\n\n- Wang 2024 (GSE184950) SNpc\n\n- Adams 2024 (GSE193688) SN\n\n- Fullard 2024 SNpc + M1 (CELLxGENE d5d0df8f-...) and tested for PD-vs-control per-donor pseudobulk differential at FDR<0.1. At least 3 of 4 atlases show log2FC > 0 in the T-cell-module score in the PD direction.\n\nH2 (secondary — region specificity within multi-region atlas): Within the Fullard atlas (the only multi-region atlas in the panel), the T-cell-module signal is significantly higher in SNpc + cingulate-like cortical regions than in DMNX (Braak 1, vagal-driven, no expected adaptive immune component) or V1 (Braak 6, late-stage, no expected ongoing immune infiltration). Pre-registered ranking: SNpc > M1 > V1 ≈ DMNX. Spearman ρ on rank vs. signature score > 0.5, permutation p < 0.05.\n\nH3 (secondary — T-cell-module × DA-loss covariation): At the donor level, in Kamath / Wang / Adams / Fullard atlases, the T-cell-module pseudobulk score is positively correlated with the AGTR1+ DA-loss score (computed identically to parkinson-da-vulnerability proposal): donors with more T-cell signal also have more vulnerable- DA loss. Spearman ρ > 0.3, permutation p < 0.05 in ≥ 2 of 4 atlases.\n\nH4 (secondary — astrocyte co-localisation in spatial slabs): Where Visium-spatial data is available (Martirosyan A5, Ma 2025 itself), spatial spots scoring high for the T-cell module are *also* enriched for CD44+ / GFAP-high reactive astrocyte markers (Ma 2025's reported co-localisation), with Spearman ρ > 0.3 on per-spot z-scored module scores.\n\nH5 (secondary — non-α-syn-homology bound — negative control): We pre-register a non-PD-relevant T-cell sub-signature (a generic exhaustion module: PDCD1, HAVCR2, LAG3, TOX, TIGIT) as a negative control. The exhaustion module should NOT show preferential PD enrichment (this signal is canonical to tumor immunology and PD T cells are not exhausted in the same way). A positive enrichment of the exhaustion module would suggest the H1 signal is a non-specific \"T cell present\" signal rather than a PD-specific clonally-expanded program.\n\nA null result on H1 is informative: it would mean that without TCR-seq ground truth, the clonal-expansion signal in Ma 2025 is not detectable from gene expression alone — calling into question any retrospective inference of adaptive immunity from existing PD snRNA-seq atlases.", "report": null, "page": "/campaign/runs/parkinson-tcell-foothold-r2/" }, { "run": "parkinson-tcell-foothold-r3", "topic": "parkinson-tcell-foothold", "attempt": 3, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 1, "partial": 0, "not_supported": 3, "inconclusive": 0, "not_testable": 1, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Ma, J., Al-Dalahmah, O., Goldman, J.E., Sulzer, D., Sims, P.A. et al. (Nat Commun s41467-025-62478-3, 2025; PMC12322057) reported spatially resolved adaptive immunity in Parkinson's disease substantia nigra: combined snRNA-seq + 10x Visium + scTCR-seq across 44 donors (13 PD + 15 ctrl SN; 10 PD + 8 ctrl cingulate; 265,284 nuclei + 10 donors with Visium) identified clonally expanded CD8+ T cells whose TCRs share homology with α-synuclein peptides and which spatially co-localise with CD44+ reactive astrocytes in PD SN.\n\nThe Ma 2025 result depends on rare TCR-seq data, which the Kamath 2022 (GSE178265), Wang 2024 (GSE184950), Adams 2024 (GSE193688) and Fullard 2024 multi-region atlases do not have. The question is whether the transcriptomic *footprint* of the Ma-2025 T-cell signature (the genes co-expressed in the clonally expanded CD8+ population) can be detected in those four independent atlases, in a deconvolution-inferred T-cell or lymphocyte cluster, without TCR ground truth.\n\nHypotheses (pre-registered):\n\nH1 (primary — transcriptomic footprint replicates in independent PD snRNA-seq atlases): The Ma 2025-defined T-cell signature (extracted from Supplementary Data of PMC12322057 — the top-50 differentially expressed genes in the clonally expanded CD8+ population vs. all other immune nuclei in PD SN) is scored per cell in T-cell / lymphocyte clusters of:\n\n- Kamath 2022 (GSE178265) SNpc + caudate\n\n- Wang 2024 (GSE184950) SNpc\n\n- Adams 2024 (GSE193688) SN\n\n- Fullard 2024 SNpc + M1 (CELLxGENE d5d0df8f-...) and tested for PD-vs-control per-donor pseudobulk differential at FDR<0.1. At least 3 of 4 atlases show log2FC > 0 in the T-cell-module score in the PD direction.\n\nH2 (secondary — region specificity within multi-region atlas): Within the Fullard atlas (the only multi-region atlas in the panel), the T-cell-module signal is significantly higher in SNpc + cingulate-like cortical regions than in DMNX (Braak 1, vagal-driven, no expected adaptive immune component) or V1 (Braak 6, late-stage, no expected ongoing immune infiltration). Pre-registered ranking: SNpc > M1 > V1 ≈ DMNX. Spearman ρ on rank vs. signature score > 0.5, permutation p < 0.05.\n\nH3 (secondary — T-cell-module × DA-loss covariation): At the donor level, in Kamath / Wang / Adams / Fullard atlases, the T-cell-module pseudobulk score is positively correlated with the AGTR1+ DA-loss score (computed identically to parkinson-da-vulnerability proposal): donors with more T-cell signal also have more vulnerable- DA loss. Spearman ρ > 0.3, permutation p < 0.05 in ≥ 2 of 4 atlases.\n\nH4 (secondary — astrocyte co-localisation in spatial slabs): Where Visium-spatial data is available (Martirosyan A5, Ma 2025 itself), spatial spots scoring high for the T-cell module are *also* enriched for CD44+ / GFAP-high reactive astrocyte markers (Ma 2025's reported co-localisation), with Spearman ρ > 0.3 on per-spot z-scored module scores.\n\nH5 (secondary — non-α-syn-homology bound — negative control): We pre-register a non-PD-relevant T-cell sub-signature (a generic exhaustion module: PDCD1, HAVCR2, LAG3, TOX, TIGIT) as a negative control. The exhaustion module should NOT show preferential PD enrichment (this signal is canonical to tumor immunology and PD T cells are not exhausted in the same way). A positive enrichment of the exhaustion module would suggest the H1 signal is a non-specific \"T cell present\" signal rather than a PD-specific clonally-expanded program.\n\nA null result on H1 is informative: it would mean that without TCR-seq ground truth, the clonal-expansion signal in Ma 2025 is not detectable from gene expression alone — calling into question any retrospective inference of adaptive immunity from existing PD snRNA-seq atlases.", "report": { "kind": "pdf", "url": "/assets/reports/parkinson-tcell-foothold-r3.pdf", "bytes": 344457, "title": "Zero pseudobulk replication of a TCR-defined T-cell signature across four Parkinson's disease snRNA-seq atlases", "main_conclusion": "Negative result: the 51-gene Ma 2025 clonal CD8+ T-cell module did not replicate (replication rate 0.0, n=2 testable atlases, delta vs shuffled null -0.555), and H2 region specificity (rho=-0.100, p=0.601), H3 T-cell/DA-loss covariation (all rho<0.3) and the H5 exhaustion negative control were also null; the report concludes TCR-seq ground truth is needed to detect clonal CD8+ programs in PD brain transcriptomes.", "thumb": "/assets/img/thumbs/parkinson-tcell-foothold-r3.webp" }, "page": "/campaign/runs/parkinson-tcell-foothold-r3/" }, { "run": "pd-fm-vs-pca-benchmark-extended-r1", "topic": "pd-fm-vs-pca-benchmark-extended", "attempt": 1, "disease": "PD", "census_class": "EARLY", "final_status": "in_progress", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": null, "report": null, "page": "/campaign/runs/pd-fm-vs-pca-benchmark-extended-r1/" }, { "run": "pd-power-floor-recoverability-r1", "topic": "pd-power-floor-recoverability", "attempt": 1, "disease": "PD", "census_class": "EARLY", "final_status": "in_progress", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": null, "report": null, "page": "/campaign/runs/pd-power-floor-recoverability-r1/" }, { "run": "pd-singlecell-reanalyzability-census-r1", "topic": "pd-singlecell-reanalyzability-census", "attempt": 1, "disease": "PD", "census_class": "EARLY", "final_status": "in_progress", "n_hypotheses": null, "counts": { "supported": 0, "partial": 0, "not_supported": 0, "inconclusive": 0, "not_testable": 0, "not_run": 0 }, "evaluated": 0, "stop_reason": "INFRASTRUCTURE", "question": null, "report": null, "page": "/campaign/runs/pd-singlecell-reanalyzability-census-r1/" }, { "run": "pd-vatpase-lysosomal-compensation-r1", "topic": "pd-vatpase-lysosomal-compensation", "attempt": 1, "disease": "PD", "census_class": "SUCCESS", "final_status": "halted", "n_hypotheses": 5, "counts": { "supported": 1, "partial": 1, "not_supported": 0, "inconclusive": 2, "not_testable": 1, "not_run": 0 }, "evaluated": 5, "stop_reason": null, "question": "Confirmatory replication of a self-generated OmiCairn finding. In a prior OmiCairn run on LRRK2 G2019S isogenic iPSC-derived dopaminergic (iPSC-DA) neurons (NanoString Human Neuropathology panel), nine CLEAR-network V-ATPase subunit genes — ATP6V0C, ATP6V0D1, ATP6V0E1, ATP6V0E2, ATP6V1A, ATP6V1D, ATP6V1E1, ATP6V1G2, ATP6V1H — were PREDOMINANTLY UPregulated in LRRK2 G2019S vs isogenic corrected controls (8/9 up at the donor level, mean log2FC = +0.485), and the direction was not comparison-specific (also +0.452, 7/9 up vs an independent healthy control). We interpret this as a COMPENSATORY transcriptional response: failing lysosomal acidification in PD-genotype DA neurons drives TFEB/CLEAR-network upregulation of the V-ATPase proton pump. The original cohort was small (NanoString, donor-level permutation power-limited), so the direction is a SUPPORTED but under-powered lab-KB hypothesis, not an established fact.\n\nThis run PRE-REGISTERS THE OBSERVED DIRECTION (V-ATPase UP) and tests it in INDEPENDENT cohorts and modalities. Cohorts: (1) GSE167335 — the LRRK2 G2019S isogenic iPSC-DA NanoString panel (independent replication of the exact original modality, where the panel covers the V-ATPase subunits it measures); (2) ArrayExpress E-MTAB-7303 (Lang\n\n2020) — iPSC-DA neurons from GBA-N370S PD patients vs controls vs idiopathic PD, bulk + single-cell RNA-seq, the GBA1-mutant arm; and\n\n(3) Smajić GSE157783 — postmortem human substantia-nigra/midbrain snRNA-seq (idiopathic PD vs control), the postmortem in-vivo arm and the blind held-out validation cohort. This pre-registers a clear honesty guard: we are testing — NOT assuming — that the compensatory UP-direction replicates; a clean reversal or null in independent data is the headline.\n\nHypotheses (pre-registered — directions are TESTED, not assumed):\n\nH1 (primary — independent LRRK2 replication of V-ATPase UP): In GSE167335 (LRRK2 G2019S isogenic iPSC-DA), the 9 measured CLEAR-network V-ATPase subunits are UPregulated in LRRK2 G2019S relative to isogenic corrected controls, replicating the prior observed +0.485 mean log2FC direction. Decision rule: donor-level (NOT replicate-level — collapse differentiation replicates to donor means to avoid pseudo-replication, exactly as the prior finding required) mean log2FC across the V-ATPase set, with a donor-label permutation test (report the minimum achievable permutation p given n) + per-gene 95% bootstrap CI. H1 REPLICATES if the mean direction is positive AND ≥6/9 subunits are individually up with the donor-level permutation p at its achievable floor. A reversal (net down) or a null is reported verbatim as failure-to-replicate.\n\nH2 (secondary — GBA1-mutant direction in an independent modality): The compensatory V-ATPase UP-direction generalises from LRRK2 to the OTHER major lysosomal-membrane PD gene, GBA1. In E-MTAB-7303 (GBA-N370S iPSC-DA vs control), the V-ATPase / CLEAR-network module is UPregulated in GBA-N370S DA neurons. Decision rule: per-sample pseudobulk (single-cell arm) and bulk DE for the V-ATPase module; report module-level effect size + FDR + 95% CI. H2 supports the compensation model if GBA-N370S shows positive module shift (FDR<0.05, CI excluding 0). The competing model (V-ATPase DOWN with failing lysosome) is equally reportable; do NOT tune toward UP.\n\nH3 (secondary — postmortem-SN direction + held-out blind validation): A V-ATPase / CLEAR-network module score, LOCKED from the iPSC-DA arms (the 9-gene set + any weights), is scored WITHOUT refitting on the held-out Smajić GSE157783 postmortem SN/midbrain atlas (idiopathic PD vs control), restricted to the dopaminergic compartment. Report the PD-vs-control module effect + 95% CI and, as a classification readout, AUC + 1000× bootstrap CI for separating PD from control donors on the module score. The lock file is frozen before touching GSE157783. A null/opposite postmortem direction is informative (iPSC compensation may not survive to end-stage postmortem tissue) and is reported as such.\n\nH4 (secondary — DA-neuron selectivity): The V-ATPase compensatory shift is SELECTIVE for the dopaminergic compartment rather than a global neuronal/glial effect. In the cohorts with multiple cell types (E-MTAB-7303 single-cell; GSE157783 SN), compare the V-ATPase module shift in DA neurons vs non-DA neurons vs glia. Decision rule: per-donor module score × cell-class interaction; report the DA-vs-other contrast effect + p + 95% CI. H4 supports selectivity if the DA shift significantly exceeds the non-DA shift. A non-selective (global) shift is an informative alternative and reported plainly.\n\nH5 (secondary — honest-failure / sensitivity + power decomposition): Decompose the replication into: (i) the achievable donor-label-permutation power floor at each cohort's n (the original limitation — state the minimum p reachable and whether the cohort can even support a significant call); (ii) sensitivity to the V-ATPase gene-set definition (the 9-gene CLEAR subset vs a broader TFEB/CLEAR target set vs V0-only vs V1-only); (iii) normalisation / batch sensitivity (NanoString housekeeping panel vs bulk vs single-cell pseudobulk). State explicitly that an under-powered null (cannot reject) is NOT the same as a replicated reversal, and report which cohorts are merely under-powered vs genuinely contradictory.\n\nA negative or partial result is fully informative. If independent cohorts show V-ATPase DOWNregulation (true lysosomal failure rather than compensation), or only an under-powered null, THAT is the headline — the self-generated compensatory-UP hypothesis must be reported as not-replicated rather than retuned (gene-set cherry-picking, replicate-level inflation) until it reads UP.", "report": { "kind": "html", "url": "/campaign/runs/pd-vatpase-lysosomal-compensation-r1/report/", "title": "Genotype-divergent lysosomal V-ATPase compensation: replicated in LRRK2 G2019S (log2FC +0.48), null in GBA1 neurons", "main_conclusion": "The 9-gene CLEAR V-ATPase module is upregulated in an independent LRRK2 G2019S iPSC-DA cohort (mean log2FC +0.4846, 8/9 up; permutation p=0.167 at the n=2+2 floor, 'REPLICATES' but not statistically confirmable). It is null in the post-hoc GBA+/N370S arm (d=-0.042) and directionally elevated in postmortem PD DA neurons (d=+5.14, degenerate CI). H4 selectivity is inconclusive and no cohort can reach p<0.05.", "thumb": "/assets/img/thumbs/pd-vatpase-lysosomal-compensation-r1.webp" }, "page": "/campaign/runs/pd-vatpase-lysosomal-compensation-r1/" } ]