Estimating the Predictive Power of Silent Mutations on Cancer Classification and Prognosis

Background Silent mutations are changes in the nucleotides comprising the DNA that do not affect the linear sequence of encoded amino acids. As their name suggests, until recent years it was believed that these mutations have no effect on the structure or function of the produced protein and thus, they are seldom used for cancer diagnosis. Therefore, this research aims to evaluate the ability of silent mutations to predict cancer types and survival probabilities over time.

Methods For cancer type classification six predictors were generated for each type. Five aim to distinguish the cancer type using a single mutation kind (Silent, Non-Silent, UTR, Intron or Flank mutations), another commingles all kinds of mutations. For survival prediction Random Survival Forest was used, based on log-rank separation of Kaplan-Meier survivability estimators.

Results Cancer type classification based solely on silent mutations achieved an F1 score at least 2.5 times higher than the F1 score expected by a null model, for all cancer types. It also outperformed the predictors using UTR, Intron or Flank mutations for some of the cancer types. Survival prediction based solely on silent mutations achieved a better AUC score than the AUC expected by the null model for eight years after the initial cancer prognosis.

Conclusions The results demonstrate that silent mutations possess important predictive power and should not be ignored. A better understanding of these mutations can both improve cancer diagnosis and cancer therapy.