Feature Fusion and Selection for Recognizing Cancer-Related Mutations from Common Polymorphisms

Single nucleotide polymorphisms (SNPs) are the most common form of genetic variant in humans, which can be generally classified into disease related mutations and common ones. It has been generally accepted that SNPs caused amino acid substitutions are of particular interest as candidates for affecting susceptibility to complex diseases, such as cancer, which is a serious public issue affecting millions of people worldwide each year. In this study, we have developed an automated and robust method to distinguish cancer-related mutations from common polymorphisms from amino acid sequence, which has a significant meaning for the cancer diagnosis, prognosis and treatment. Multiple different sequential features are extracted and the most important features are finally selected for constructing the prediction model. Experimental results show that an overall 81.07% success rate has been obtained, indicating the proposed method is very promising in the clinical cancer research studies.