Predicting Protein Structural Class Based on Ensemble Binary Classification

With resent advances in deriving protein sequence´s features, a new feature has been proposed for the purpose of enhancing the prediction quality, which named Quasi-sequence-order. These descriptors are derived from both the Schneider-Wrede physicochemical distance matrix and the Grantham chemical distance matrix between each pair of the 20 amino acids. The feature was taken as the input of a new designed neural network to develop statistical learning models for predicting the protein structural class. The statistical model is designed to reduce four classes to six binary classifiers, and then by ensemble the six classifiers to get the final prediction. It was experiment through the rigorous jackknife cross validation text that the success rates by our method were significantly improved.