Protein secondary structure prediction using optimal local protein structure and support vector machine

Abstract

Protein includes many substances, such as enzymes, hormones and antibodies that are necessary for the organisms. These proteins have different shapes and structures which distinct them from each other. By having unique structures, only proteins able to carried out their function efficiently. The importance of understanding protein structure has fueled the development of protein structure databases and prediction tools. The main objective of this research is to optimize local protein structure with Support Vector Machine (SVM) to predict protein secondary structure. Most of the related study used fixed segment length for secondary structure prediction and this might produce inaccurate results. In this research, dataset is segmented into different segment length of local protein structure. An optimal length of local protein structure is determined and the evaluation is carried out by comparing with the existing methods and initial prediction using native structure. Higher accuracy and true positive rate, low false positive rate are obtained which prove the effectiveness of this prediction method. A statistical method, t-test, is applied to validate the results of the prediction.