Exploring the gut microbiome in type 2 diabetes across different insulin resistance levels: a machine learning approach

Insulin resistance (IR) sits at the heart of type 2 diabetes mellitus (T2DM), and composite indices such as atherogenic index of plasma (AIP), the metabolic score for insulin resistance (METS-IR), the triglyceride-glucose index (TyG) and TyG-BMI are widely used to evaluate its severity. To explore the link between IR and the gut microbiome (GM), we collected blood test outcomes and stool samples, sequenced 16S rRNA genes from the stools, trained XGBoost models to differentiate individuals with high IR from healthy controls, and conducted correlation analyses among GM features, clinical measures and IR indices. The results showed that TG, FBG, and HDL-C levels differed significantly between participants with T2DM and healthy controls. Indices of IR were also markedly higher in the T2DM group. Machine learning analysis of GM profiles demonstrated high performance in distinguishing T2DM individuals with high IR indices from healthy controls, with GM features including Bacteroides and Faecalibacterium contributing most to the XGBoost models. Correlation analyses further revealed that GM features including Lachnospiraceae_UCG-010, Bacteroides, Faecalibacterium, Lachnospira, Parasutterella, and Escherichia-Shigella are associated with clinical measures and IR indices. These findings highlight these GM features as potential intervention targets for improving insulin resistance and restoring carbohydrate and lipid metabolism.