General Framework for Two-stage Approaches to Protein Secondary Structure Prediction
Bioinformatics techniques to protein secondary structure prediction, such as Support Vector Machine (SVM) and GOR approaches, are mostly single-stage approaches; they predict secondary structures of the protein by taking into account only the information available in amino acid sequences. On the other hand, PHD (Profile network from HeiDelberg) method is a two-stage technique where two Multi-Layer Perceptrons (MLPs) are cascaded; the second neural network receives the output of the first neural network captures any contextual relationships among the secondary structure elements predicted by the first neural network. In this paper, we argue that it is feasible to extend the current single-stage approaches by adding a second-stage prediction scheme to capture the contextual information among secondary structural elements and thereby improving their accuracies. We demonstrate that two-stage SVMs perform better than present techniques for protein secondary structure prediction.