A deep bidirectional long short-term memory approach applied to the protein secondary structure prediction problem

Abstract

One of the most important open problems in science is the protein secondary structures prediction from the protein sequence of amino acids. This work presents an application of Deep Recurrent Neural Network with Bidirectional Long Short-Term Memory (DBLSTM) cells to this problem. We compare the performance of the proposed approach with the state-of-the-art approaches. Despite the lower complexity of the proposed approach (i.e. Neural Network architecture with fewer neurons), results showed that the DBLSTM could achieve a satisfactory level of accuracy when compared with the state-of-the-art approaches. We also studied the behavior of Gradient Optimizers applied to the DBLSTM. Furthermore, this paper concentrates on well-known quantitative analytical methods applied to evaluate the proposed approach.