Computational Prediction Of Driver Missense Mutations In Melanoma

Discovering driver mutations used as the diagnostic and prognostic biomarkers is important for the treatment of cancer, including melanoma. Although during the last decade several computational methods have been developed to predict the effect of missense mutations in cancer, only a few have been specifically designed for identifying driver mutations in a specific disease context. To take into consideration of disease-specific factor, here we made efforts to prioritize missense mutations presented in melanoma. We collected 385 pathogenic mutations from the database of curated mutations (DoCM), and 392 benign mutations filtered from a benchmark neutral database (VariSnp), respectively. To evaluation of the model effect, we also selected 45 mutations from other databases. Then a random forest classifier was constructed to prioritize melanoma pathogenic mutations based on conservation, functional region annotation, protein secondary structure, protein domain, physicochemical features, and splicing information. The proposed method achieved an AUC of 0.94 on both training and test sets. When compared with previous developed algorithms, our method obtained a higher accuracy in identifying driver missense mutations in melanoma, along with a more balanced sensitivity and specificity than the other prediction methods.