Microbial Signatures of Disease: Classifying Type 2 Diabetes with Microbiome Data in a US Cohort

Ekaterina Degtiareva , University of Oxford
Charles Rahal, University of Oxford
Andrea Tilstra, University of Oxford
Jennifer Dowd, University of Oxford

This study investigates whether microbiome data can enhance the classification of type 2 diabetes (T2D) beyond traditional risk factors in a population-based setting. We use data from the National Longitudinal Study of Adolescent to Adult Health (Add Health), a nationally representative US cohort now linked with 16S rRNA gut microbiome data. As a first step, we generate and describe a set of microbiome-derived features—including diversity metrics, taxonomic abundances, and functional profiles—across 1,384 individuals. We then compare the performance of a benchmark logistic regression model to a set of machine learning classifiers (LASSO, random forest, XGBoost, and SVM), each trained with and without microbiome features. Models are evaluated using AUROC, F1 score, sensitivity, specificity, and calibration. This study contributes to emerging research on the microbiome as a socially patterned biological system and tests its potential utility in improving the classification of metabolic disease.

See extended abstract

 Presented in Session 51. Health, Morbidity and Wellbeing