An Effective Feature Extraction Method on Protein Secondary Structure Class Prediction

摘要

Knowledge of protein structures plays a significant role in fields like protein function analysis, pharmaceutical research, and drug design. Extracting pivotal and representative features is the basis of the secondary structure class prediction of proteins. Over the past few decades, scientists mainly focused on low-similarity protein sequences. Although many different measures of feature extraction have been found for low-similarity data sets, the prediction accuracy still remains to be upgraded. In this paper, a new feature extraction method is proposed, which integrates statistic features, pre-dicted secondary structure segments, approximate entropy and hydrophobicity patterns of protein sequences. The prediction accuracy is obtained by the jackknife test, based on three widely used low-similarity benchmark datasets (25PDB, 1189 and 640). Compared with other existing methods, the method that proposed in this paper has achieved the highest overall accuracy.