Abstract:
Large-scale genome sequencing efforts have generated somatic mutation data across large cohorts of patients and many tumor types. This comprehensive collection of data has enabled a more complete understanding of the mutational landscape in various cancers. One of the key challenges for researchers going forward is to distinguish the pathogenic driver mutations that lead to cancer from the neutral passenger mutations that do not directly contribute to the disease phenotype. We developed a novel deep learning method that uses a convolutional neural network to model the effects of somatic mutations on protein structure and stability to identify driver mutations in cancer. The CNN model accurately identified driver and passenger mutations from large-scale sequencing projects. It outperformed traditional machine learning methods and many popular effect predictors. The model could prove to be a useful tool for researchers in the search for driver mutations that play an important role in cancer initiation and progression and might help to understand the mechanisms of oncogenesis.
