Nan Shen, Nevenka Dimitrova, Cleo H. Ho, Pedro J. Torres, Francine R Camacho, Ying Cai, Momchilo Vuyisich, Damon Tanton, Guruduth Banavar
Recognizing and treating the early stages of type 2 diabetes (T2D) is the most cost effective way to decrease prevalence, before heart disease, renal disease, blindness, and limb amputation become inevitable. In this study, we employ high resolution gut microbiome metatranscriptomic analysis of stool samples from 53,970 individuals to identify predictive biomarkers of type 2 diabetes progression and potential for diagnosis and treatment response. The richness of the metatranscriptomic data enabled us to develop a T2D risk model to delineate individuals with glycemic dysregulation from those within normal glucose levels, with ROC-AUC of 0.83+/-0.04. This risk score can predict the probability of having insulin dysregulation before detecting high glycated hemoglobin (HbA1c), the standard-of-care marker for prediabetes and diabetes. Additionally, a machine learning model was able to distinguish novel metatranscriptomic features that segregate patients who receive metformin and are able to control their HbA1c from those who do not. These discoveries set the stage for developing multiple therapeutic avenues for prevention and treatment of T2D.