Protein secondary structure prediction based on integration of CNN and LSTM model

Abstract

Protein structure prediction is an important issue in computational biology, and protein secondary structure prediction is the basis for protein three-dimensional structure prediction. A protein secondary structure prediction method based on convolutional neural networks (CNN) and Long Short-Term Memory (LSTM) is proposed in this paper. The architecture of CNN has two convolutional layers, one max-pooling layer and one ReLU activation layer. The feature maps extracted from second convolutional layer are used to feed to softmax classifier, and the first probability output is obtained. The LSTM model has a sequence layer and a last layer. The feature is extracted from last layer and input to random forest classifier to get the second probability output. The two probabilistic outputs are weighted and integrated to obtain the prediction model EN-CSLR in this paper. Based on the advantages of integration of the two models, cross-validation experiments are performed on the 25pdb dataset, and Q3 reaches 80.18%, which is higher than using only one model. The experimental results show that the features extracted from CNN and LSTM models can effectively improve the accuracy of protein secondary structure prediction.