Classification of G protein-coupled receptors using attention mechanism

In order to get a better understanding of G Protein-Coupled Receptors (GPCRs) biological function, a precise classification of this kind of receptors in the databases is a real need. To improve the classification accuracy, machine learning algorithms are encountered the major challenge such as the extraction of valuable features. In this paper, we introduce AttentionFam method, which utilizes a novel deep learning architecture to overcome the challenges of prevailing approaches, which are domain specific and computationally intensive. The AttentionFam employs advantages of attention mechanism and representation learning to represent implicitly the features of both the aligned and unaligned GPCR protein sequences. Therefore, feature extraction was carried out from raw protein sequences and thus no sequence alignment methods such as MSA are needed. To evaluate the proposed approach, an extensive set of experiments conducted. The results showed that our proposed method achieved the good accuracy of 97.40%, compared to the state-of-the-art approaches. In addition, it showed better performance in terms of time consumption and less memory for the same data analysis.