Prediction of Protein Structural Classes for Low-Similarity Sequences Based On Predicted Secondary Structure

Abstract

Knowledge about structural classes of proteins plays an important role in inferring tertiary structure and function of a protein. One of the major problems with the existing algorithm for the prediction of protein structural classes is low accuracies for proteins from + and / classes. To improve accuracies, one needs to extract features with high representation power. Several authors proposed enormous number of features. Some of them redundant, most of them overlapping. In this paper, most prominent features proposed in the literature are reviewed. Features extracted from Position Specific Scoring Matrices (PSSM) are excluded and left as the subject matter of another paper. Also some combinations of these features are used to classify a low-homology dataset, 25PDB, and 30FB, with sequence similarity lower than 25% and 30%, respectively. Comparison of our results with others shows that to find the best combination is very important and may provide a cost-effective alternative to predict protein structural class in particular for low-similarity datasets.