Generative AI-Augmented Proteomics for Predicting and Linking Cerebrovascular Diseases and Cognitive Decline
Description: Vascular contributions to cognitive impairment and dementia (VCID) may account for up to 50% of dementia cases worldwide and reflect a complex interplay between cerebrovascular disease (CeVD) and cognitive decline. Identifying individuals at risk across both conditions and uncovering the molecular signatures they share are critical steps toward prevention and targeted intervention for cerebrovascular contributors to cognitive decline. Plasma proteins, as downstream products of gene-environment interactions, offer a promising window into pathobiology underpinning both CeVD and cognitive decline. Biobank-scale resources such as the UK Biobank Pharma Proteomics Project (UKB-PPP) now enable profiling of thousands of circulating proteins across tens of thousands of individuals. However, prior studies often model disease outcomes separately, missing cross-disease proteomic patterns that may reflect shared risk profiles and mechanisms across diseases. We propose to apply (1) multi-outcome machine learning models to improve protein-based prediction of outcomes across both CeVD and cognitive outcomes, and (2) graph representation learning to reveal shared protein signatures linked to established mechanisms such as blood-brain barrier dysfunction, inflammation, and endothelial injury as well as potentially novel pathways. In our preliminary studies, a linear multi-outcome prediction model using the UKB-PPP proteomics panel outperformed single-outcome baselines across seven major CeVD and cognitive outcomes, and an exploratory graph representation analysis revealed proteins aligned with neuroinflammatory pathways positioned near both stroke and cognitive decline in a joint protein-disease embedding space. Beyond using the UKB-PPP data, we will also leverage generative artificial intelligence (GenAI) to synthesize validated protein abundances for all UK Biobank participants and assess whether GenAI-augmented data preserves or improves predictive accuracy and biological interpretability. By integrating machine learning, GenAI, and population-scale proteomics, this project leverages shared proteomic variation for risk prediction and mechanistic discovery across CeVD and cognitive decline. Results of this project are likely to support biomarker development and inform precision strategies to target VCID and prevent cognitive decline linked to vascular disease.
Funder: American Heart Association
Amount: $70,680
Funding type: training
Funder scheme: Predoctoral Fellowship
Start date: 2026-07-01
End date: 2028-06-30
Investigator: Zhiyu Yan (lead)
Institutions: Harvard University, President and Fellows of Harvard College, Harvard T. H. Chan School of Public Health
Subfield: Psychiatry and Mental health
Field: Medicine
Domain: Health Sciences
Funded outputs count: 0
Funder award ID: 26PRE1563536
OpenAlex ID: G520582953
