Prediction protein structural classes with a hybrid feature

Select the proper feature of protein sequence is a crucial step in protein structural class prediction. In this paper we intend to propose a novel hybrid feature to describe the protein. This hybrid feature is composed of two parts, one is physicochemical composition (PCC), and another is the recurrence quantification analysis (RQA). A new classifier is constructed with the Error Correcting Output Coding (ECOC) which incorporates three binary Artificial Neural Network (ANN) classifiers. We select 1189 data set to verify the efficiency of classify. The accuracy of our method on this data set is 57.3%, higher than some other methods on the same datasets. Furthermore only 33 parameters are used in our method, lower than many other methods. This indicates that the hybrid feature we proposed here is promising to the prediction of protein structural classes.