Fig. 5

Comparative analysis predictive model performance across all metabolites in the targeted dataset for both 6-weeks (n = 158) and 12-months (n = 282) time points. Top panel shows superimposed boxplots and violin plots of the distribution of predictive posterior mean for each evaluation metric across all 36 metabolites. Bottom panels show aggregated model rankings for all metabolites using R-squared (left) and Spearman correlation (right) using Borda scores (Methods). Higher scores indicate that a model was consistently selected as a better performing. Relatively similar Borda scores and cross-metabolite average predictive performances indicate that no model was clearly the most performant. However, support vector machines (with radial basis function kernel) was highest scoring model