Type 2 Diabetes Subtyping via Phenotype and Genotype Co-Learning.

Area of Science:

• Genomics
• Biomedical Informatics
• Endocrinology

Background:

• Type 2 Diabetes (T2D) interpretation and subtyping are critical for understanding pathophysiology and clinical stratification.
• Existing methods often use limited pre-selected factors, hindering comprehensive analysis of multimodality data.

Purpose of the Study:

• To develop and validate a novel hypergraph framework for integrated analysis of phenotypic and genetic data in T2D.
• To achieve fine-grained pathophysiological insights and precise clinical stratification for T2D patients.

Main Methods:

• Utilized a cohort of 42,256 participants from the National Institutes of Health's All of Us Research Program.
• Developed a hypergraph framework integrating phenotypic and genetic features for T2D prediction and subtyping.
• Employed end-to-end clustering of clinical concepts, genetic variants, and individuals.

Main Results:

• The hypergraph framework achieved an Area Under the Receiver Operating Characteristic curve (AUROC) of 89.64% in predicting T2D.
• The pipeline successfully performed subtyping by clustering diverse data types.
• Analysis of genetic risk scores identified distinct genetic profiles among T2D subtypes.

Conclusions:

• The integrated hypergraph approach offers a powerful tool for T2D prediction and subtyping.
• The findings underscore the potential for precision medicine applications in managing T2D.
• This method advances the comprehensive analysis of multimodality data in complex diseases.