Deep Learning Solutions for Protein: Recent Development and Future Directions

Abstract

AbstractThe prediction of protein structure has a major role in drugs design and network pharmacology. However, complexity of the protein structure, time consumption and expensive cost incurred by the state-of-the-art methods of predicting protein structure motivated researchers to propose deep learning solutions. Despite the growing applications of deep learning in protein structure, no dedicated comprehensive survey on the exploration of deep learning in protein structure. In this paper, a comprehensive survey on the exploration of deep learning in protein structure is presented. This survey starts by addressing a quantitative and synthesis analysis to show an insight on deep learning in protein structure prediction. Based on the synthesis analysis provided, the survey created a taxonomy of the literature related to exploring deep learning in protein structure. Research challenges and new perspective for future research direction for developing deep learning solutions for protein structure prediction are discussed in details. Specifically, the survey identified that the exploration of nature inspired algorithms in deep learning for protein structure remain untapped. In addition, hybrid/ensemble deep learning architecture is receiving unprecedented attention and growing in number at an exponential rate. In another twist, no single contributor from Africa contributed to the development of this research area. We believed that our survey will served as a starting point for novice researchers and it can help expert researchers to quickly identify research direction for developing deep learning solutions for exploring protein structure.KeywordsDeep learningConvolutional neural networkStacked auto-encoderDeep belief networkGenerative adversarial networkProtein structureGenomeProtein interaction