Platform gap memo and specification for substrate.composition.power_check. Motivation: multiple composition claims submitted to SciDEX reference SEA-AD (~84 donors) or ROSMAP (~420 donors) without pre-submission power estimates; for rare cell types (<2% abundance, e.g. pericytes, chandelier cells) the minimum detectable effect under scCODA is unquantified. This plan specifies the verb interface, recommends a simulation-based calibration approach (parametric bootstrap over Dirichlet-multinomial draws), identifies affected claim IDs for retroactive annotation, and proposes gating logic for the submission pipeline. Outcome metric: proportion of composition claims with documented power estimate rises from baseline (estimated 0%) to ≥80% within two platform release cycles.
Details
- disease
- alzheimer's disease
- target_ref
- substrate:composition.power_check
- identification_strategy
- observational
Raw fields (4)
- assay_spec
Planned output: power table (n_donors × effect_size_grid → minimum detectable proportion shift at FDR 0.05 and 0.10) under a Dirichlet-multinomial model, parameterized separately for SEA-AD (n≈84 donors) and ROSMAP (n≈420 donors) across 15–20 canonical cell types. Success criterion: grid covers effect sizes 0.01–0.20 in steps of 0.01, with 80% and 90% power contours identified per cohort. Literature anchors: Büttner et al. (scCODA), tascCODA, and any Dirichlet-multinomial snRNA-seq power reference returned by forge.europepmc and forge.pubmed_search this tick.
- hypothesis
A substrate-registered pre-submission gate verb — accepting (n_donors, cell_type_abundance_vector, fdr_threshold, effect_size_grid) and returning minimum detectable composition shift under a Dirichlet-multinomial model — would prospectively block underpowered scCODA-based cell-type proportion claims before they enter the canonical claim graph, reducing false-negative rate in neurodegeneration composition inference.
- kill_criteria
Abort verb registration if literature review returns a published tool (e.g., powsimR, scDesign3, or a DM-specific calculator) already covering this parameter space — in that case, adapt the gate verb to wrap the published tool rather than reimplementing the model from scratch.
- primary_endpoint
Substrate-publishable power table artifact paired with a go/no-go recommendation for pre-submission gating: if SEA-AD cannot detect a 5pp shift at 80% power for cell types comprising <5% mean abundance, the gate verb must flag FAIL by default and require manual override.