A 9-state hidden Markov model using protein secondary structure information for protein fold recognition

Abstract

In protein fold recognition, the main disadvantage of hidden Markov models (HMMs) is the employment of large-scale model architectures which require large data sets and high computational resources for training. Also, HMMs must consider sequential information about secondary structures of proteins, to improve prediction performance and reduce model parameters. Therefore, we propose a novel method for protein fold recognition based on a hidden Markov model, called a 9-state HMM. The method can (i) reduce the number of states using secondary structure information about proteins for each fold and (ii) recognize protein folds more accurately than other HMMs.

Introduction

The genome projects provided sequence information for a large number of proteins. An accurate 3D structural representation of proteins is necessary for determining detailed functionalities of these sequences [1]. Among methods for determination of protein structures is the experimental method, such as X-ray crystallography and NMR [2]. However, the experimental determination of protein structures remains a long and difficult task. To contend with these problems, many international researchers are developing structure prediction methods [1], [3], [4]. Protein structure prediction methods are used to predict 3D structures and various properties, secondary structures and domain boundaries of proteins from protein primary structures. Protein 3D structure prediction detects the first homology between protein sequences [3]. In the case of high sequence similarity, finding homologies is very easy. According to similarities of proteins, tertiary structure prediction is divided into three steps as follows [4]: first, when the similarity of two sequences is very high, fragment based comparative modeling easily finds homologies between protein sequences. Second, the fold recognition method is used to find remote homologies or distant homologies. Third, when template proteins among known 3D structure proteins cannot be found, the New Fold method, called ab initio or de novo, is used.

A newly identified protein can be related to proteins in annotated databases for which the 3D structure is known. The knowledge of protein structural class such as fold assists in the determination of the 3D structure of a protein. If the structural class of a protein is known, it can be used to considerably reduce the search space of structure prediction processes, since most of the structure alternatives could be eliminated and therefore the structure prediction task is simplified and the whole process is accelerated [10].

There are many tools for protein tertiary structure prediction using hidden Markov Models (HMMs) [3], [5], [6], [7], [8], [9], [10], [11]. In sequence-based approaches, HMMs are predominant and also demonstrate high performance [5], [11]. HMMs were applied for multi-class protein fold recognition [6], employing sequence alignment and modeling (SAM) software [5]. Secondary structure information was incorporated into HMMs [7]. Karchin et al. [9] used an identical approach and they evaluated different alphabets for backbone geometry. Douchaffra et al. applied relations between components (e.g. secondary structures of proteins) of an entity and the whole [8]. Lampros proposed a reduced state-space HMM, which simultaneously finds amino acid sequences and secondary structures for proteins [10]. However, the main disadvantage of HMMs is the employment of large model architectures, which require large data sets and high computational resources for training. It is necessary to reduce the parameters of HMMs, while HMMs maintain performance of fold recognition. Also, HMMs must consider sequential information about secondary structures of proteins, to improve prediction performance.

Therefore, we propose a novel fold recognition method for protein tertiary structure prediction based on a hidden Markov model. The method consists of nine hidden states and uses protein sequences and secondary structure information to recognize the fold of proteins. Our contribution can be summarized as follows: first, we present the 9-state HMM suited to protein fold recognition. The nine states are divided into three components: $\alpha$-helix, $\beta$-strand, and coil. Each state of the three components represents the “beginning”, “middle”, and “ending” for the secondary structure. Second, we evaluate the proposed HMM for a low homology data set. Finally, we conjecture that the performance of protein fold recognition can be improved using amino acid frequency and secondary structure information in fold recognition. We also avoid utilizing the computationally expensive Baum–Welch algorithm [21] and reduce the number of parameters for HMM.

The rest of this paper is organized as follows: Section 2 introduces related work about protein structure prediction. In Section 3, we describe the proposed hidden Markov model. Section 4 shows a set of test data and several experimental results. Finally, Section 5 summarizes our conclusion and future work.