Prediction of protein secondary structure using SVM-PSSM Classifier combined by sequence features

Protein secondary structure prediction is an important problem in bioinformatics. For this task, a method based on the SVM-PSSM Classifier combined by sequence feature (SF) is proposed in this paper. Protein sequence data is represented by a hybrid formation which combines the Position-Specific Scoring Matrix (PSSM) with the Hydrophobicity Sequence Feature (HSF), and the Structural Sequence Feature(SSF). The novel feature formulations are proposed to extract more than 40 SFs(sequence features), which further are examined for feature selection as accessional input vectors to SVM-PSSM classifier, to construct more accurate classifiers. The prediction classifier in this study is constructed using PSSM and selected sequence features. The experimental results show that our final constructed SVM-PSSM-SF classifier achieved accuracy Q
₃
of 78%, that is higher than the traditional method which using the amino acid sequence as the data set.