Decoding Alzheimer’s Amyloid-Beta Pathology through Astrocytic and Microglial GRNs via Single-Cell AI

Overview

Alzheimer's disease (AD) affects over 55 million individuals globally, yet fewer than 5% of cases are diagnosed at the mild cognitive impairment stage when therapeutic intervention is most effective, and an estimated 40–60% of synapses are already lost before clinical symptom onset. Conventional differential gene expression (DGE) methods analyze genes in isolation, failing to capture the coordinated regulatory programs driving this early neurodegeneration. To address this gap, I developed a network-informed biomarker discovery framework integrating gene-regulatory connectivity with single-cell expression profiles from the AD Progression Atlas (32 donors; astrocytes and microglia; 5 brain regions; quantitative pTau and Thal amyloid staging). Gene regulatory networks were inferred using ARACNe and integrated with GREmLN to generate regulatory embeddings, upon which region-specific classifiers were trained under strict patient-level cross-validation. Network-informed models achieved a 2-fold improvement in F1 score over DGE-based classifiers (0.69 ± 0.08 vs. 0.35 ± 0.05; p < 0.01), and perturbation-based feature importance identified 5,663 biomarker candidates significantly enriched for synaptic maintenance (adjusted p = 3.2 × 10⁻⁶) and neuroinflammatory pathways (adjusted p = 1.7 × 10⁻⁴), with AGRN and DVL1 emerging as convergent regulators of synaptic-integrity disruption. Independent validation in the SEA-AD cohort (84 donors; middle temporal gyrus) reproduced this advantage, with embedding classifiers outperforming expression-only models by 75% in F1 score (0.42 vs. 0.24). These findings demonstrate that incorporating regulatory network structure into representation learning significantly enhances early-stage AD classification, offering a scalable paradigm for transcriptomic biomarker discovery across complex neurodegenerative diseases.

Competition history

• CSEF 2026 Computational Science (Senior Division) · Entry S-07-08

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Computational Sciences projects
• CSEF Computational Sciences projects
• Every CSEF 2026 project