Fig. 1

Nested cross validation. To implement nested cross validation, we split the data set into ten stratified folds. The figure shows one out of ten NCV loops. For each NCV iteration, an independent testing set (F ⁽¹⁰⁾ in the figure) is left out to assess FATHMM-indel’s performance. The remaining folds (red sets in the figure) are merged to create the tuning set used to learn, under cross validation, the optimal values of the hyperparameters. Crucially, a different fold is used as testing set in each iteration, fully exploiting all data to evaluate FATHMM-indel’s performance