Neuroimaging multi-omics fingerprints for subtyping and uncertainty-aware prediction of alzheimer's disease

The National Institute on Aging awarded $526,607 to Wake Forest University on September 1, 2026, under the Aging Research program (CFDA 93.866) to develop computational imaging frameworks that integrate neuroimaging with multi-omics data to identify Alzheimer's disease subtypes and model neurodegenerative progression with calibrated uncertainty. The project, titled "Neuroimaging Multi-Omics Fingerprints for Subtyping and Uncertainty-Aware Prediction of Alzheimer's Disease," will construct knowledge-guided graphs to characterize multimodal imaging-omics interactions and identify interpretable AD subtypes, and design uncertainty-aware deep learning models to forecast neurodegenerative and cognitive trajectories. The work uses ANMERGE data for model development and discovery, with generalizability evaluation on complementary datasets (AMP-AD and ADNI). The award funds research performed in Winston-Salem, North Carolina, with a period of performance extending through August 31, 2029. Successful completion will yield novel deep learning algorithms that unify multi-omics and neuroimaging data to reveal interpretable subtypes and generate uncertainty-aware disease predictions.

Abstract

Alzheimer's disease (ad) is a leading cause of dementia worldwide, yet its progression is highly heterogeneous across individuals. A major limitation in biomedical imaging is the inability to connect neuroimaging phenotypes to their underlying molecular drivers, which limits the biological interpretability and clinical predictive power of neuroimages. While neuroimaging genetics has linked genetic variants to brain structure, IT overlooks critical regulatory layers (transcriptomic and proteomic processes) that mediate the path from gene to brain. Moreover, most predictive models for ad progression output a single trajectory without quantifying prediction confidence, making IT difficult to translate into clinical decision-making. The overall objective of this project is to develop computational imaging frameworks that integrate neuroimag- ING with multi-omics data (genomics, transcriptomics, and proteomics) to identify ad subtypes and model NEU- rodegenerative progression with calibrated uncertainty. We hypothesize that neuroimaging multi-omics finger- prints capture the molecular-imaging relationships that explain phenotypic heterogeneity and improve the relia- bility of individualized progression predictions. Aim 1 constructs a knowledge-guided graph to characterize multi- modal imaging-omics interactions and identify interpretable ad subtypes. Aim 2 designs uncertainty-aware deep learning models to forecast neurodegenerative and cognitive trajectories. The project will use anmerge data for model development and discovery, and evaluate generalizability on complementary data (AMP-ad, adni). Successful completion of this project will yield novel deep learning algorithms that unify multi-omics and NEU- roimaging data to reveal interpretable subtypes and generate uncertainty-aware disease predictions. The out- comes will advance precision neuroimaging and medicine by providing generalizable computational frameworks applicable across alzheimer's and other neurodegenerative diseases. Undergraduate students will participate in data integration, method implementation and evaluation, and result interpretation under close mentorship, gaining hands-on research experience at the intersection of neuroscience and artificial intelligence.