Optimal approach to sequence-to-sequence prediction: applications in bioinformatics

We propose a two-stage approach to sequence-to-sequence prediction problem by using support vector machines (SVMs) to optimize the prediction from single stage techniques. The sequence-to-sequence prediction problem is common to many bioinformatics applications and we demonstrate our approach by using it on the protein secondary structure prediction problem. The new predictor combining different types of GOR (Gamier, Osguthorpe, and Robson) and Bayesian classifiers with SVMs achieves an accuracy of 70.9% when using the sevenfold cross validation on a database of 126 nonhomologous globular proteins. Extending the method to multiple sequence alignments of homologous proteins significantly increases the prediction accuracy to 72.1%. The results show that it is possible to obtain a higher accuracy with combined hierarchical classifiers than single stage classifiers alone, in the sequence prediction.