Using support vector machines to distinguish enzymes: Approached by incorporating wavelet transform

Abstract

The enzymatic attributes of newly found protein sequences are usually determined either by biochemical analysis of eukaryotic and prokaryotic genomes or by microarray chips. These experimental methods are both time-consuming and costly. With the explosion of protein sequences registered in the databanks, it is highly desirable to develop an automated method to identify whether a given new sequence belongs to enzyme or non-enzyme. The discrete wavelet transform (DWT) and support vector machine (SVM) have been used in this study for distinguishing enzyme structures from non-enzymes. The networks have been trained and tested on two datasets of proteins with different wavelet basis functions, decomposition scales and hydrophobicity data types. Maximum accuracy has been obtained using SVM with a wavelet function of Bior2.4, a decomposition scale j=5, and Kyte–Doolittle hydrophobicity scales. The results obtained by the self-consistency test, jackknife test and independent dataset test are encouraging, which indicates that the proposed method can be employed as a useful assistant technique for distinguishing enzymes from non-enzymes.

Introduction

Enzyme is one of the most important biological catalysts in the metabolism of all organisms. Although the information about enzyme structure can be determined by conducting various experimental methods such as biochemical analysis of eukaryotic and prokaryotic genomes or microarray chips, which are both expensive and time-consuming (Chou and Elrod, 2003). Particularly, the number of newly found protein sequences has increased explosively in the post genomic era. For instance, in 1986 Swiss-Prot contained only 3939 protein sequence entries, but in 2008 the number has jumped to 385721 according to version 55.4 of the UniProtKB/Swiss-Prot Release as of May 20 2008, implying that the number of protein sequences has increased by about 55 times in about two decades. Facing such a “protein sequence explosion”, it is both challenging and indispensable to develop an automated method for fast and reliably annotating the enzyme attributes of uncharacterized proteins. The knowledge thus obtained can help us timely utilize these newly found protein sequences for research.

We aim to demonstrate that protein function can be predicted as enzymatic or not. Determination of a given new sequence belonging to enzyme or non-enzyme remains vital and challenging task using bioinformatics tools. In the past, a number of algorithms have been developed for this problem including sequence similarity (Baxevanis, 1998; Bork and Koonin, 1998; Schuler, 1998), evolutionary analysis (Eisen, 1998; Benner et al., 2000), hidden Markov models (Fujiwara and Asogawa, 2002), structural consideration (Teichmann et al., 2001; Di Gennaro et al., 2001), protein/gene fusion (Enright et al., 1999; Marcotte et al., 1999), protein interaction (Aravind, 2000; Bock and Gough, 2001), motifs (Hodges and Tsai, 2002), neural-networks (Fujiwara and Asogawa, 2002; Jensen et al., 2002a), and family classification by sequence clustering (Enright and Ozounis, 2000; Enright et al., 2002). In the absence of clear sequence or structural similarities, the criteria for comparison of distantly-related proteins become increasingly difficult to formulate (Enright and Ozounis, 2000). Moreover, not all homologous proteins have analogous functions (Benner et al., 2000). The presence of the shared domain within a group of proteins does not necessarily imply that these proteins perform the same function (Henikoff et al., 1997). Many proteins sharing promiscuous domains are known to have very different functions (Marcotte et al., 1999). Therefore, the development of new independent methodologies that are able to distinguish enzymes from protein sequences with advantages in some aspects can complement current methods and will further strengthen the ability to computationally predict enzyme structures.

In order to solve the problems mentioned above, a lot of modifications have been produced in the past few years. Support vector machine (SVM) based on amino acid sequence is one of them (Dobson and Doig, 2003, Dobson and Doig, 2005; Chou, 2005; Zhou et al., 2007). SVM is well-founded theoretically on statistical learning theory, and has many attractive features including effective avoidance of over-fitting, ability of handle large feature space and absence of local minima. However, as a machine learning technique, SVM requires a fixed length of pattern, it is not possible to use this technique in case of protein with too small or too large length (Sudipto and Raghava, 2006). Introduced in the early 1980s, wavelets have become a popular signal analysis tool due to their ability to elucidate simultaneously both spectral and temporal information within the signal. Wavelet transform (WT) is a local time-frequency analysis method with both changeable time window and frequency window. Because of its character of multi-resolution, WT has been applied in bioinformatics to analyze and process biological data recently (Lio, 2003; Mandell et al., 1997a; Mandell et al., 1998; Selz et al., 1998; Giuliani et al., 2000; Selz et al., 2004; Selz et al., 2007; Mandell et al., 1997b; Qiu et al., 2004, Qiu et al., 2003; Lu et al., 2004).

This paper is devoted to combining the discrete wavelet transform (DWT) based on the physicochemical property of residues and SVM to develop a new predictor for distinguishing enzyme structures from non-enzymes. Maximum accuracy has been obtained by the new method with biorthogonal 2.4 (Bior2.4) of decomposition scale j=5 for Kyte–Doolittle hydrophobicity scales (KDHΦ). It was observed that the accuracy using combined method was higher than WT or SVM only.