Classification of Protein sequences using hybrid Recurrent Deep Learning Models

It is evident that protein databases are enriched with over several million entries of protein sequences with a significantly wide variety of individual protein functional domains. To add a significant contribution towards this research path, we propose supervised learning-based hybrid recurrent deep learning models by combining Convolutional Neural Network and with Long Short Term Memory (CNN-LSTM) and Gated Recurrent Unit (CNN-GRU) into the learning architecture for protein sequence classification task with better performance. The proposed hybrid models are equipped to extract the features of neighboring amino acids with CNN module and extract sequential information of amino acid sequence with LSTM or GRU modules. The hybrid CNN-LSTM and CNN-GRU models exhibit high performance in terms of high training accuracy score of 97.5 %. The proposed model exhibits high performance comparatively with a high Receiver Operating Curve (ROC) Area Under Curve (AUC) score of 0.995 with less computational complexity elapsing for 2100 seconds to complete training process.