What does a scalar PRS hide?
A reproducible public-reference workflow maps signed Alzheimer genetic effects through declared variant-to-gene mappings and GTEx tissue/context expression to individualized biological allocation profiles. The objective is not to replace scalar PRS, but to measure what it compresses away and which higher-level structure survives reasonable representation changes.
Similar aggregate risk can conceal different inferred biological allocations.
Pathway-resolved scoring was extended with explicit decomposition, compression audit, representation challenge, and refusal controls. The purpose is to ask which structure remains defensible after plausible changes in summary, centering, annotation, and coordinate assumptions.
Promote only what survives the analyst’s choices.
Signed allocation profiles
Map signed GWAS effects through declared gene and tissue/context representations into individual allocation profiles.
Make cancellation visible
Separate centered risk-positive and protective contributions rather than silently absorbing them into a net score.
Retain residual structure
Measure how much higher-dimensional allocation is explained by scalar PRS and preserve what remains unresolved.
Replace plausible annotations
Freeze an alternative biological vocabulary and test whether the higher-level structure persists.
Deny hostile inputs promotion
Use weak or permissive coordinate constructions as controls and assign them no positive mechanistic weight when they fail.
Require held-out value
Phenotype claims remain closed until retained features improve a prespecified held-out or external decision.
This is a structure result, not a phenotype result.
No Alzheimer phenotype, treatment response, donor-tissue archetype, or causal vascular result is established by this analysis. The stable grouping is a representation-robust descriptive result. Incremental phenotype value remains open and requires held-out or external validation.
When does allocation become a quantitative phenotype?
Which retained allocation features add reproducible information beyond scalar PRS under held-out phenotype prediction, donor stratification, or mechanistically targeted experimental design?
Incremental value remains an empirical question for held-out or external validation; it is not inferred from the descriptive structure alone.