Proteome-Wide Association and LLM-Based Prioritization of Type 2 Diabetes Protein Targets

Type 2 diabetes is a major cause of illness and health disparities. Genetic association studies can identify regions of the genome associated with disease risk, but they do not always identify the proteins or biological mechanisms that contribute to disease development. This project will conduct a retrospective secondary analysis of existing, controlled-access genetic, proteomic, and phenotype data, including data accessed through the UK Biobank and other previously collected datasets. No new participants will be recruited, enrolled, contacted, treated, or followed as part of this study.

The study will use proteome-wide association methods to evaluate whether genetically predicted circulating protein levels are associated with type 2 diabetes risk. Analyses will consider evidence across African American and European American datasets when available, with attention to population-specific and shared signals. Statistical genetic evidence will be integrated with relevant biological and clinical information to prioritize protein targets that may have a causal role in type 2 diabetes.

Large language model-based methods will be used as a structured evidence-synthesis tool to organize and summarize publicly available information relevant to prioritized proteins, including biological function, disease relevance, and potential therapeutic tractability. All computational results will be reviewed by the research team. The project will generate reproducible analytic workflows, a ranked list of candidate protein targets, and hypotheses for future experimental validation. Findings are intended for research use and will not be used to make clinical decisions for individual patients.

No new participants will be recruited for this study. The study will conduct a retrospective secondary analysis of existing, controlled-access datasets. Eligible records are from adult participants aged 40 to 84 years at enrollment in the source studies who have available genetic data and plasma proteomic data for population-specific protein prediction modeling. Participants with and without type 2 diabetes may be included, depending on the source dataset and analytic objective. Type 2 diabetes genome-wide association summary statistics will also be used; no individual-level participant contact or enrollment will occur.