Protein Secondary Structure Prediction Using Ensemble of LSTM Neural Networks

The prediction of protein structure plays an important role in understanding protein structure composition and protein biological function, which is an important part of protein structure prediction. The characteristics of biological information carried by protein residues are the prerequisites for predicting protein secondary structure. After the position specific evolution matrix is adopted to encode a residue, each residue can be represented as a two-dimensional data plane. In this paper, the Long and Short Memory Network(LSTM) is used as a training method, the length and width of the two-dimensional data plane can be respectively used as the time dimension of an LSTM to design two LSTM models, and then the two models can be combined to design a third LSTM model. Finally, the three models were integrated with the ensemble method, and the Q3 result obtained by the ensemble method in the CB513 protein dataset was 77.9.