Figure 1.

Feature importance for insulin- and glucometabolic markers based on extreme gradient boosting models and linear regression for the most important predictors. Panel (A) to Panel (F) displays relative importance for predictors generated by extreme gradient boosting models, using several pre-processing techniques for metabolomics data to further reduce number of predictors in the final machine-learning model. Model diagnostics (RMSE) and validation (R²) are presented next to each prediction model. For each outcome, the most important predictors identified through machine learning were included in linear regression models. All regression models were adjusted for age. Significance level are described as follows: *p-value < 0.05, **p-value < 0.005, ***p-value < 0.0005. p-values < 0.095. < denotes that lower levels for the predictor was associated with the target variable. > denotes that higher levels for the predictor was associated with the target variable.