Combining evolution and protein language models for an interpretable cancer driver mutation prediction with D2Deep (Updated October 29, 2024).

The article discusses the challenges in distinguishing cancer-causing driver mutations from random passenger mutations and introduces the AI-based D2Deep method to address this issue. D2Deep combines a non-specialized protein language model with protein-specific evolutionary information to outperform existing predictors and offer result interpretation. The model, trained on a balanced somatic training set, helps mitigate biases related to hotspot mutations and can be used for mutation prioritization in clinical settings. The preprint emphasizes the importance of unbiased predictive models in cancer research and provides a link for accessing D2Deep predictions and confidence scores for clinical interpretation. [Extracted from the article]