Cross-validation

Evaluation of CHAMOIS against N=1,598 BGCs with known metabolites from MIBiG 3.1. Model performance on every predicted ChemOnt class (N=539) using stratified grouped 5-fold cross-validation. Classes are displayed as nodes within the ChemOnt hierarchy and coloured according to CHAMOIS’s performance for the respective class, as assessed by the area under the precision-recall curve (AUPRC, see colour key). Precision-recall curves for selected classes are shown as insets against the baseline obtained by random guessing (dashed horizontal lines corresponding to class proportions).