Single-cell machine learning uncovers genetically anchored, cell-type specific programs of Alzheimer's disease.

TriSCOPE applied to ~4.2 million nuclei from 627 donors identifies cell-type-specific predictively expressed genes in Alzheimer’s disease that are largely non-overlapping with univariate differential expression.

• Why it matters: Univariate marginal testing in single-cell studies cannot resolve multivariate disease structure or separate clinical from neuropathological signals, limiting genetic and cell-type interpretation.
• What they did: The framework combines multivariate predictive modeling, donor-aware differential expression, and pseudotime inference across six snRNA-seq datasets, then performs conditional GWAS-eQTL colocalization to link genes to risk loci.
• The result: It yields 27 genetically supported gene–cell type pairs (including ARL17B, JAZF1) with no significant differential expression, nominates convergent programs, and shows clinical–neuropathological divergence plus cross-cohort replication of ARL17B, LINGO1, and APOE.

AI summary of the abstract · may contain errors · not medical advice · read the paper.