Fig. 6: Comprehensive lifestyle prediction of metabolic subphenotypes and validation on independent cohort.

a Diet, sleep and physical activity correlation network analysis. Concurrent correlations between lifestyle features were calculated using Spearman correlation with permutation and considered significant if multiple testing-adjusted q value < 0.2. The color gradation represents correlation coefficients ranging from −1.0 to 1.0. Different colors of points indicate different types of lifestyle features: light green (diet); purple (sleep); red (activity); dark green (combined features from diet and sleep); blue (combined features from diet, sleep, and activity). b–g Integrated lifestyle prediction model of metabolic subphenotypes. The LASSO classification model was built upon all lifestyle features, and model coefficients of selected features were visualized. The classifications are for (b). Normoglycemia vs. PreDM/T2D. c Incretin normal vs. dysfunction; (d) beta-cell normal vs. dysfunction; (e) muscle IS vs. IR; (f) adipose IS vs IR; and g Hepatic IS vs. IR. Different colors indicate different types of lifestyle features. Sex (1 male, 0 female) and ethnicity (1 Caucasian, 0 non-Caucasian) are two levels of numerical values. Latency is the time spent to accomplish the transition from full wakefulness to sleep onset. %E Meal, energy proportion (%) of the meal timing to the total daily energy intake; %Carb, carbohydrate proportion (%) of the food group out of the total daily carbohydrate intake from all food groups; movement/sedentary duration, the ratio of movement duration to sedentary duration; education, the years of education, PreDM prediabetes, IS insulin sensitive, IR insulin resistance. Nutrients (e.g., fiber and sodium) are the daily dietary intakes of the corresponding nutrients. h Comparison of accuracy for predicting incretin function in training main cohort and validation cohort. The Y axis indicates classification accuracy (%).