Type 2 Diabetes Subtyping via Phenotype and Genotype Co-Learning.
Journal:
Studies in health technology and informatics
Published Date:
Aug 7, 2025
Abstract
Interpreting and subtyping type 2 diabetes (T2D) is challenging yet essential for achieving fine-grained pathophysiological insights and precise clinical stratification. Previous studies have primarily relied on a small number of pre-selected risk factors and biomarkers, neglecting the integration of multimodality data (e.g., phenotypic and genetic features) for more comprehensive analyses. In this study, we select a cohort of 42,256 participants from the National Institutes of Health's All of Us Research Program, where our hypergraph framework achieves an AUROC of 89.64% on predicting T2D when integrating phenotypic and genetic features. The proposed pipeline performs subtyping by clustering clinical concepts, genetic variants, and individuals in an end-to-end manner. Further analysis using genetic risk scores reveals distinct genetic profiles between T2D subtypes and highlights the potential applications of our solution in precision medicine.
