NeuroAxisNet: Brain-Blood Network-Guided Transformers Reveal Early Signatures of Alzheimer’s Progression and Resilience

Overview

Alzheimer’s disease (AD) affects 1 in 9 Americans over age 65, yet 98% of clinical trials fail, reflecting a mismatch between cellular heterogeneity and coarse-grained patient-level single-target therapies. To identify early biomarkers and cell-type-specific therapeutic targets, I developed NeuroAxisNet, the first computational framework to model Alzheimer’s disease as a continuous progression axis using multi-tissue data and identify multi-target biomarkers distinguishing resilient from degenerative trajectories. NeuroAxisNet introduces a network-guided transformer framework that integrates single-cell, spatial, and peripheral blood transcriptomes from 9 datasets (364 donors; >950K cells across neurons, microglia, astrocytes, and blood). By incorporating gene regulatory structure, the model generates denoised cell representations that resolve distinct biological pathways, such as synaptic signaling and metabolism. Diffusion pseudotime orders cells along a continuous disease progression axis. When validated across independent spatial datasets, NeuroAxisNet outperformed existing methods, achieving 91.07% kNN module purity (vs. 83.92% for standard transformer and 75.14% for graph neural networks) and producing NeuroAxis scores strongly aligned with neuropathology (Spearman ρ = 0.83 vs. 0.64 for PCA). Importantly, cells from the same donor, cell type, and pathological stage diverged into resilient and degenerative trajectories. Differential analysis identified both established AD-associated genes (APOE, DLG4/PSD95, RAB3A) and novel candidates (RASGEF1C, LINGO1, RASGEF1B, SLC17A7), with therapeutically relevant hubs (GPNMB, HLA-DRA/B2M), forming a coordinated biomarker panel, with several detectable in peripheral blood for non-invasive early detection. Ultimately, NeuroAxisNet provides a timing-aware, multi-tissue framework for early diagnosis and target prioritization, helping address the high failure rate of current Alzheimer’s therapies.