Protein Secondary Structure Prediction Based on Deep Learning

Abstract

Prediction of protein secondary structure from the amino acid sequence is a classical bioinformatics problem in computational biology. For accurate predicting the sequence-structure mapping relationship between protein secondary structure and features, a novel deep learning prediction model is proposed by combining convolutional neural network (CNN) and bi-directional recurrent neural network (BRNN) with long short-term memory cells (Bi-directional LSTM RNN). In order to draw eight classes (Q8) protein secondary structure prediction results, we first utilize CNN to filter and sample amino acid sequences, and then use Bi-directional LSTM RNN to model context information interaction between amino acids in protein. Experimental results show that the prediction accuracy of the proposed model is about 1-3% higher than that of the existing prediction models, and the prediction accuracy of 69.4% is obtained.