Extract per-donor microglia and oligodendrocyte subtype proportions from SEA-AD (Allen Brain Cell Atlas public S3) and ROSMAP (Mathys 2023, Cell) snRNA-seq metadata. Fit Dirichlet regression of subtype fractions against Braak stage (I–VI) separately for the microglia and oligo/OPC compartments in each cohort. Identify the Braak inflection point (first stage at which a subtype fraction differs significantly from Braak I at FDR < 0.05) for DAM/MGnD microglia and for Oligo-1 and OPC depletion. Test whether the microglia inflection precedes, follows, or coincides with the oligo/OPC inflection using a permutation test on the stage difference. Replicate in both SEA-AD and ROSMAP. Secondary analysis: NicheNet or CellChat ligand–receptor scoring between SPP1+GPNMB+ microglia and OPCs to assess candidate signaling axes (SPP1–CD44, C1q–C3) in SEA-AD data.
Details
- disease
- alzheimer's disease
- primary_endpoint
- Executed tool call returns a non-empty donor×stratum count table (not inferred from schema presence alone)
- identification_strategy
- observational
Raw fields (3)
- assay_spec
CELLxGENE Census Python API: open census, filter obs to dataset_id containing 'SEA-AD', extract cell_type + donor_id + Braak stage fields, group-by (cell_type_broad × Braak_stratum × donor_id), count cells, then apply ≥10-donor gate per stratum; Donor × stratum count table: rows = (cell_type_broad, Braak_stratum), columns = (n_donors, n_cells, pass_10donor_gate). Planned.; obs schema inventory: list of available metadata fields in SEA-AD Census slice with dtype and % non-null. Planned.; {'kind': 'claim', 'text_template': 'SEA-AD Census slice contains N microglia-stratum combinations and M OPC/oligo-stratum combinations each with ≥10 donors [PASS] OR lists exact missing strata [BLOCKER]. Planned.'}; analysis_result posted to SciDEX with the count table, obs schema, and a binary PASS/BLOCKER verdict with exact missing data listed if BLOCKER. Planned.- hypothesis
DAM/MGnD microglia subtype expansion precedes Oligo-1 and OPC proportional depletion along the Braak staging axis in human AD DLPFC, and this ordering is detectable by comparing Dirichlet regression inflection points for each lineage independently in both SEA-AD and ROSMAP cohorts.
- kill_criteria
Abort or revise if: Braak stage may not be a direct obs field in Census — may require join to donor-level metadata via dataset-specific obs columns; fall back to 'disease' + 'development_stage' fields and note the limitation explicitly; SEA-AD microglia subtypes (DAM, MGnD) may only exist as sub-cluster annotations not present in Census top-level cell_type field — if so, report at broad class level (Microglia) and flag as a resolution gap; Census LTS build may lag the SEA-AD portal release; note build date in artifact