Single-cell analysis maps Alzheimer’s phenotypes to brain cell subpopulations

Researchers developed an artificial intelligence–based framework to investigate cellular correlates of the diverse clinical phenotypes of Alzheimer’s disease (AD). The study used the PsychAD single-nucleus RNA sequencing resource, comprising more than 6 million nuclei from the prefrontal cortex of over 1,000 brains. The Phenotype Associated Single Cell encoder (PASCode) combines statistical methods in a graph neural model to score associations between individual cells and phenotypes.

The analysis identified approximately 1.5 million phenotype-associated cells from 584 donors with AD-related phenotypes. Comparisons across 27 brain cell subclasses prioritized microglial subpopulations associated with AD pathology and reactive astrocyte subtypes potentially linked to cognitive resilience. Cognitively impaired AD donors showed patterns consistent with increased excitatory/inhibitory imbalance and mitochondrial dysfunction. Astrocytes associated with both AD and depression displayed gene-expression patterns involving inflammation and endoplasmic reticulum stress pathways.

The authors validated findings in external population-scale datasets covering AD and major depressive disorder. They also produced an AD phenotypic single-cell atlas and released PASCode as an open-source tool with pretrained models and a web application. These results may help prioritize diagnostic and therapeutic research targets, but the observational associations do not establish causality or demonstrate clinical utility. The supplied abstract does not report effect sizes or quantitative validation performance.