A novel feature extraction method by compressive sensing for signal peptide

Cui-Fang Gao, Qiang Guan, Hao

Abstract

In this paper, we propose a novel mathematical expression of signal peptide that can truly reflect the intrinsic attributes of the sequence. To deal with every signal peptide that displays diverse length, we first transformed the original sequence into Markov transition matrix with a unified size. Next we obtained the expansion sparse vector on a unit orthogonal basis, and finally utilized random projection of compressive sensing (CS) to obtain an accurate feature expression. Analyzing from the perspective of traditional mathematical theory, we determined that the feature vector abstracted by CS was the optimal combination of the original information (amino acid composition, sequence order, and so on). Thus, the new method can be regarded as a comprehensive development of high-density discriminative information extraction. The experimental results suggested that the CS feature expression has the approving determinant and has the potential for future research and applications in the development of feature extraction.