Application of deep learning methods for protein structure prediction

Abstract

The rapid development of machine learning and deep learning methods has led to their widespread use in many areas of research, including their use in solving problems of proteomics. Today, machine learning methods play an important role in predicting the three-dimensional structural components of proteins by interpreting how protein sequences and their homology govern interresidual contacts and structural organization. Recent CASPs (Critical Assessment of Structure Prediction) assessing the state of the art in modeling protein structure based on amino acid sequence have demonstrated significant progress in modeling structures without the use of structural templates (historically “ab initio” modeling). The progress was driven by the successful application of deep learning techniques to predict the distances between residuals. In turn, these results led to a significant increase in the accuracy of the three-dimensional structure, provided that a sufficient number of sequences are known for a family of proteins. In addition, the number of sequences required for alignment has been significantly reduced, and the accuracy of templatebased models has also improved significantly. This paper provides an overview of recent advances in the application of deep learning methods used to predict the three-dimensional structure of a protein. The possibilities of using neural networks to identify unknown protein structures and functions of proteins, which is one of the most important tasks of proteomics, are also considered. Problems that have yet to be solved are described, but it is expected that in the near future the described methods will play a decisive role in the structural bioinformatics of proteins.