Applying expression profile similarity for discovery of patient-specific functional mutations

Abstract

The progress of cancer genome sequencing projects yields unprecedented information of mutations for numerous patients. However, the complexity of mutation profiles of patients hinders the further understanding of mechanisms of oncogenesis. One basic question is how to uncover mutations with functional impacts. In this work, we introduce a computational method to predict functional somatic mutations for each of patient by integrating mutation recurrence with similarity of expression profiles of patients. With this method, the functional mutations are determined by checking the mutation enrichment among a group of patients with similar expression profiles. We applied this method to three cancer types and identified the functional mutations. Comparison of the predictions for three cancer types suggested that most of the functional mutations were cancer-type-specific with one exception to p53. By checking prediction results, we found that our method effectively filtered non-functional mutations resulting from large protein sizes. In addition, this methods can also perform functional annotation to each patient to describe their association with signalling pathways or biological processes. In breast cancer, we predicted \"cell adhesion\" and other mutated gene associated terms to be significantly enriched among patients.