Fig. 4

Feature importance for the two best performing mutation classification models that utilized TIP and TMEC features. a Feature importance for the PDGFRB gene mutation classifier (logistic regression). Here, feature importance is measured by the value of its regression coefficient. b Feature importance for the RET gene mutation classifier (random forest). Here, the importance is measured by the reduction of the Gini index obtained when the feature is added to the tree, averaged across the trees in the random forest model. c Distribution of feature values for four of the most important TIP or TMEC features, as presented in (a), divided between patients with the mutated and non-mutated PDGFRB gene. d Distribution of feature values for four of the most important TMEC features, as presented in (b), divided between patients with the mutated and non-mutated RET gene