Precision phenotyping of type 2 diabetes in chinese populations using a variational autoencoder-informed tree model

Area of Science:

• Endocrinology and Metabolism
• Genetics and Population Studies
• Computational Biology and Bioinformatics

Background:

• Type 2 diabetes (T2D) presents significant clinical heterogeneity, complicating management.
• Current T2D classification models, primarily based on European cohorts, demonstrate limited generalizability to other ancestries.
• Addressing population-specific T2D heterogeneity is crucial for advancing precision diabetology.

Purpose of the Study:

• To evaluate the generalizability of existing T2D classification models across different ancestries.
• To develop and validate a novel, population-specific classification framework for T2D in a Chinese cohort.
• To identify key clinical features driving T2D heterogeneity in Chinese individuals.

Main Methods:

• A tree-like graph structure from Scottish data was tested on a large, multi-center Chinese cohort (32,501 patients).
• A variational autoencoder (VAE) framework was employed to identify key clinical features in the Chinese cohort.
• The Discriminative Dimensionality Reduction Tree (DDRTree) algorithm was used to construct a Chinese-specific T2D tree model, validated in external cohorts.

Main Results:

• While cardiovascular and kidney outcomes showed similar distributions, diabetic retinopathy varied across ancestries within similar phenotypes.
• The developed Chinese T2D tree model captured population-specific heterogeneity.
• Longitudinal analysis revealed phenotypic shifts within the Chinese cohort trending towards higher-risk branches of the classification tree.

Conclusions:

• Existing T2D classification frameworks require adaptation for diverse populations due to ancestry-specific phenotypic variations.
• A population-specific classification framework, like the one developed for the Chinese cohort, is essential for accurate T2D risk prediction.
• Implementing tailored classification systems will facilitate individualized risk stratification and specialized treatment guidelines in precision diabetology.