Fig 3. Baseline metabolites predict trained immunity responses 3 months after BCG vaccination.

(A) The dataset was randomly split into a training set (70% of the observations) and a test set (30%). Then, a 10-fold cross-validation was performed for feature selection, which was repeated 5 times, to generate a prediction model for IL-1β, IL-6, and TNF-α fold changes in response to S. aureus 3 months after BCG vaccination. Prediction accuracy was evaluated by calculation of a Pearson correlation between the measured cytokine fold changes and the predictions of the test sets. This procedure was repeated 100 times, and the correlation coefficients are visualized as boxplots. (B–D) The top 20 metabolites occurring in the prediction models for IL-1β (B), IL-6 (C), and TNF-α (D) fold changes in response to S. aureus 3 months after BCG vaccination are visualized. In total, 100 models were constructed for each cytokine fold change. (E) Variance explained by metabolites (top 10 most often occurring metabolites from the prediction analysis), age, sex, and BMI on trained immunity responses. Values indicate the variance explained by metabolites. The metadata on study participants, metabolome, and cytokine data used to generate these figures are available at https://gitlab.com/xavier-lab-computation/public/bcg300. BCG, Bacillus Calmette–Guérin; BMI, body mass index.