Institute of Biomedical Sciences, Academia Sinica, Taipei, Taiwan

摘要

Consensus-based protein structure prediction methods have been proved to be successful in recent CASPs (Critical Assessment of Structure Prediction). By combining several weaker individual servers, a meta server tends to generate better predictions than any individual server In this paper, we present a Support Vector Machines (SVM) regression-based consensus method for protein fold recognition, which is a key component for high-throughput protein structure prediction and protein functional annotation Our SVM model extracts the features from a predicted structural model by comparing it to other models generated by all the individual servers and then predicts the quality of this model Experimental results on several LiveBench data sets show that our consensus method consistently performs better than individual servers Based on this approach, we have developed a meta server, Alignment by Consensus Estimator (ACE), which is participating in CASP6 and CAFASP4 (Fourth Critical Assessment of Fully Automated Structure Prediction). ACE is available at http://www cs uwaterloocaTByu/consensus htm.