A fast and efficient nearest neighbor method for protein secondary structure prediction

Using PSSM profiles, various machine learning methods have been successfully developed for protein secondary structure prediction. With the steady increase of protein structure data, the probability of having available homologous structural information of the protein in real prediction is now fairly high and will continue to increase. Therefore, how to effectively utilize the ever-growing protein structure data has become a huge challenge and opportunity. In this paper, we propose a novel nearest neighbor method, DPred, to use both homologous and non-homologous information for protein secondary structure prediction. On the dataset composed of new solved proteins, the method achieves the overall Q₃ and SOV scores of 87.51% and 86.50%, which is comparable with Porter_H and better than PROTEUS and CDM.