Prediction of Protein Secondary Structure Using LIBSVM and SSVM with different Kernel Function

Author Name(s): *Rajnish K. Vashishtha, Abhay banker, Neetu Verma, C. K. Verma , Manoj Jha
Author Email: rajnishmanit@gmail.com

Abstract

Bioinformatics and computational biology is the emerging field of research in current era. Proteomics is solitary of the prime areas of research in bioinformatics. The general function of proteins is mostly dictated by structure. Prediction of secondary structure plays a key role in its further prediction of tertiary structure. Several techniques have been proposed by the researchers for protein secondary structure prediction. Since the data is inherently complex and these approaches endure the problem of accuracy. There are lots of methods based on machine learning procedure, such as neural networks (NN), clustering, decision tree and LIBSVM etc. Support Vector Machine has exposed physically powerful generalization capability of prediction of protein structure. In this paper, the important point on protein secondary structure by binary classification using LIBSVM and SSVM with the diverse kernel role has been detailed. The proposed method gives a better accuracy than existing technique for the prediction of protein structure. This method can be performed in MATLAB software. Two datasets are used for this problem optimization.

Keywords

Protein Secondary Structure Prediction, Zika Virus Datasets, Support Vector Machine, Smoothing and Kernel Function

Introduction

Bioinformatics and data mining is the highest rising area of research. Proteomics is one of the richest area in Bioinformatics because protein plays significant task in a biological process, understanding the function, discovery of novel medicine and products of industrial and medical application. Many researchers attempt to determine the construction of proteins using nuclear magnetic resonance and X-ray crystallography. Both methods are time consuming & expensive [1]. Prior information on class of protein structure will significantly pick up the quality and piece of secondary structure prediction of protein from amino acid succession by dipping the search space of the structure prediction process. The idea of protein structural class was first introduced by Chortia and Levitt [2]. In general protein structure can be classified into α, β, α/β, α+β. A protein is formed by the specific sequence of amino acids. This linear string is called the primary structure. The protein secondary structure help to resolve of tertiary structure by fold recognition method. Main essential parts of secondary structure are coil or turn, α- helix, β-sheets. Third level in structure of protein is multimeric and monomeric protein molecules. It’s very difficult to predict. In this paper we are center on α- helix, β-sheets for binary categorization of protein secondary structure. Our objective is to travel around the problem of Protein structure forecast using machine learning techniques.

The key idea of machine learning is intend mechanism to learn like a human, study from knowledge and discover information from dataset. Various machine learning technique are used to handle prediction of protein structure problem in bioinformatics [3]. The study attempted to concern support vector machine and smooth support vector machine using a linear, quadratic, polynomial and RBF kernel functions for binary classification of protein secondary structure. These methods are applied on dataset. These dataset sequences are derived from NCBI & use Chou fasman algorithms. As a result smooth support vector machine reached high accuracy, this shows that the structure class of a protein is considerably correlated with its amino acid composition. The smooth support vector machine can become a helpful for forecast the structural classes of protein.

Conclusion

This research work is based on the forecasting of protein structure, and is useful for drug designing, drug discovery and data testing. All the prediction results are much closer to the different kernel functions. Also the error calculated to achieve SSVM is lesser than the error values obtained in LIBSVM approach. The SSVM approach takes less number of iterations to reach up to desired value and SSVM approach is significantly faster and provides better accuracy than LIBSVM approach. The SSVM can work with extensive range of data where input values are given with respective reference output values. Unlike LIBSVM, SSVM do not need a big training of data set. A small dataset is sufficient to train in SSVM and it becomes ready for prediction.

153 total views, 3 views today

Download PDF File

About the author: dev