Novel PSSM-Based Approaches for Gene Identification Using Support Vector Machine

Abstract

By understanding the function of each protein encoded in genome, the molecular mechanism of the cell can be recognized. In genome annotation field, several methods or techniques have been developed to locate or predict the patterns of genes in genome sequence. However, recognizing corresponding gene of a given protein sequence using conventional tools is inherently complicated and error prone. This paper first focuses on the issue of gene prediction and its challenges. The authors then present a novel method for identifying genes that involves a two-step process. First the research presents new features extracted from protein sequences using a position specific scoring matrix (PSSM). The PSSM profiles are converted into uniform numeric representation. Then, a new structured approach has been applied on PSSM vector which uses a decision tree-based technique for obtaining rules. Finally, the rules of single class are joined together to form a matrix which is then given as an input to SVM for classification purpose. The rules derived from algorithm correspond to genes. The authors also introduce another approach for predicting genes based on PSSM using SVM. Both the methods have been implemented on genome DNAset dataset. Empirical evaluation shows that PSSM based SAFARI approach produces better results.