Fig. 8

Prediction of gene expression levels based on intron and exon features. Random Forest Performance and Feature Importance. (a) Tenfold cross-validated ROC curves for Random Forests trained with intron-only, exon-only features, and both sets combined for the upper/lower quartile data set. (b) MDA feature importance for a Random Forest model trained with combined exon and intron features for the upper/lower quartile expression data set. (c) SHAP summary plot containing the 20 features with highest importance of the Random Forest model trained with combined exon and intron features for lower and upper quartile dataset. Exon features were extracted from the respective first exons of genes, as were intron-features extracted from first introns