Position Specific Scoring Matrix and Synergistic Multiclass SVM for Identification of Genes

Authors: Wani, M.A., Bhat, H.F. and Jan, T.R.

Journal: Proceedings - 17th IEEE International Conference on Machine Learning and Applications, ICMLA 2018

Pages: 1192-1196

ISBN: 9781538668047

DOI: 10.1109/ICMLA.2018.00192

Abstract:

In genome annotation field several methods have been developed to locate the patterns of genes in genome sequence. In this paper we propose a novel method for identifying genes from protein sequences. The first step of the proposed method involves computing a position specific scoring matrix (PSSM) of protein sequences. The normalized PSSM is used to convert protein sequences into training data set. The data set is simplified by averaging the normalized values corresponding to the same amino acid that occur at more than one location of the protein sequence. The resulting training data set is used to train multiclass Support Vector Machine (SVM) classifier that relates simplified normalized values of amino acid to various genes. The results of several multiclass SVMs are synergistically combined for improving the results. The proposed approach is tested on genome DNAset dataset. Empirical evaluation shows that the proposed new approach produces good results of identifying genes using protein sequences.

Source: Scopus