Fig. 7

A Comparing the best-performing model (i.e., highest mean AUPR relative to permuted baseline) trained on a single data type against the best “multi-omics” model for each target gene. None of the differences between single-omics and multi-omics models were statistically significant using paired-sample Wilcoxon tests across cross-validation folds, for a threshold of 0.05. B–G Classifier performance, relative to baseline with permuted labels, for mutation prediction models trained on various combinations of data types. Each panel shows performance for one of the six target genes; box plots show performance distribution over 8 evaluation sets (4 cross-validation folds × 2 replicates).