Grain protein function prediction based on improved FCN and bidirectional LSTM

Abstract

Abstract Predicting grain protein function from amino acid sequences is becoming more and more significant, especially with the speed at which sequencing technology is developing. Most models suffer from lower accuracy in predicting protein activity due to their neglect of the sequence order of amino acids. Therefore, A parallel PBiLSTM-FCN algorithm is proposed for predicting grain protein function. In order to further increase the model's prediction accuracy, the PBiLSTM-FCN algorithm combines the Fully Convolutional Networks (FCN) and the bidirectional Long Short-Term Memory network (BiLSTM). It also adds the Squeeze-Excitation block to the FCN algorithm's complete convolutional block. This experimental dataset includes the protein data of four grains: soybean, maize, indica, and japonica. The study results show that compared with other algorithms, the PBiLSTM-FCN algorithm can achieve better performance in predicting grain protein function, proving the effectiveness and superiority of the PBiLSTM-FCN algorithm. Furthermore, the verified and formally genuine protein functions found in the SwissProt database were compared with the experimentally anticipated functions of four significant grain proteins in this work.