Disentangling context-specific Alzheimer’s Disease progression signatures through an explainable AI framework

Disentangling context-specific Alzheimer's Disease progression signatures through an explainable AI framework

This project was completed with the Lee lab during rotations.

Disease progression is often nonlinear and stage-dependent, shaped by interacting biological pathways that evolve over the disease course. Resolving these stage-dependent, context-specific transcriptional programs requires single-cell resolution. Single-cell RNA sequencing provides this granularity, profiling diverse brain cell populations to uncover cell type-specific disease signatures, but the technology now outpaces our ability to interpret the data it produces. These datasets are high-dimensional and heterogeneous, carrying technical noise, batch effects, and complex biological covariates that can obscure subtle disease signals.

To address these challenges, we introduce ADVISE (Alzheimer’s Disease Variational Isolation of Signal Embeddings), an AI-based framework designed to capture AD progression-associated transcriptional variation in single-cell RNA sequencing datasets. This approach enables the identification of stage-dependent genes and pathways at global, cell type-specific, and sex-specific resolutions.

In this study, we apply ADVISE to large-scale single-nucleus RNA sequencing datasets to resolve transcriptional programs associated with AD progression across global, cell type-specific, and sex-specific contexts. We identify gene signatures and pathways shared across biological contexts, as well as mechanisms that are uniquely context-dependent.

In collaboration with Dr. Jessica Young and Dr. Chris Arian, we validate our candidate gene findings through CRISPR-mediated genetic perturbations in excitatory neurons and identify candidate genes for future therapies.