How Strong Is Your Driver? Machine Learning and Structural Features Predict Driver Mutation Strength and Clinical Outcomes

We present STRIDE (STRuctural-Informed Driver Evaluation), a machine learning framework that classifies cancer driver mutations into strong and weak categories using protein structural and evolutionary features. Moving beyond traditional binary driver/passenger labels, STRIDE refines driver annotation by distinguishing mutations with differential oncogenic impact. The optimized model achieved an AUROC of 0.89 on independent test data and was further validated through functional, pharmacological, and clinical analyses across multiple datasets.

Keywords: Cancer genomics, driver mutations, mutation strength classification, structural biology, machine learning.