The prevalence of pre-diabetes and type 2 diabetes mellitus (T2DM) is high worldwide, which attracts the attention of the World Health Organization. However, it is difficult to realize the early diagnosis according to blood glucose and hemoglobin A1c. Through metabolomics combined with machine learning analysis, we screened 10 biomarkers related to T2DM and impaired fasting glucose (IFG). Finally, based on eXtreme Gradient Boosting (XGBoost) algorithm, integrated biomarkers profiling is constructed with 10 biomarkers, which are mainly related to arginine and proline metabolism; valine, leucine and isoleucine degradation as well as biosynthesis; and phenylalanine metabolism. Integrated biomarkers profiling is used to predict IFG and T2DM based on metabolites panel in vivo, which provides an important reference value for early diagnosis of IFG and T2DM.

Integrated biomarkers profiling is constructed based on XGBoost algorithm, in which the concentrations of biomarkers are set as covariates and the disease status is set as dependent variable. After inputting the biomarker concentrations, the unknown sample is predicted according to the predicted value calculated by XGBoost algorithm. It could be interpreted that the unknown sample belongs to the group that has the highest predicted value in the four groups.