Abstract: Objective To develop predictive models for the risk of gestational diabetes mellitus(GDM)based on targeted metabolomics of serum bile acids. Methods Within a prospective birth cohort,serum levels of 32 bile acids were quantified in early pregnancy.Biomarkers for GDM were screened by the Least Absolute Shrinkage and Selection Operator(LASSO).The eXtreme Gradient Boosting(XGBoost)algorithm was used to build the predictive models,and bootstrap resampling was applied to evaluate model performance.The SHapley Additive exPlanations(SHAP)method was utilized to assess the importance and predictive value of biomarkers. Results The bile acid metabolomics-based models demonstrated strong performance(AUC ≥ 0.840),which outperformed models based on traditional risk factors(AUC=0.828).After incorporating traditional risk factors,predictive performance improved further(AUC ≥ 0.885).Key bile acids identified included glycoursodeoxycholic acid-3-sulfate(GUDCA-3S),taurodeoxycholic acid-3-sulfate(TDCA-3S),taurolithocholic acid(TLCA),glycohyocholic acid(GHCA),ursodeoxycholic acid(UDCA),and the ratios of glycocholic acid to cholic acid(GCA∶CA)and cholic acid to chenodeoxycholic acid(CA∶CDCA).Elevated levels of GUDCA-3S and TDCA-3S,as well as increased GCA∶CA and CA∶CDCA ratios,were associated with higher GDM risk.In contrast,higher concentrations of TLCA,GHCA,and UDCA were associated with lower GDM risk. Conclusion The bile acid metabolomics-based predictive models exhibit robust predictive efficacy.The identified bile acids may serve as potential biomarkers,providing support for early GDM identification,metabolic regulation,and mechanistic research.

Key words: birth cohort, gestational diabetes mellitus, bile acid, predictive model, SHAP