A Review of Prediction Methods of Interaction Sites of Antibody-Protein Complexes Based on Artificial Intelligence

Abstract

Introduction: Cancer is one of the most important health issues in the current and next centuries. Understanding the mechanism of interaction between antibody-protein residues is essential for designing targeted anticancer drugs based on monoclonal antibodies. Prediction of the effective structure is the first step for production of monoclonal antibodies. Methods: This paper is a systematic review of the state-of-the-art researches on prediction of interaction sites and specification of antibody structures. Artificial neural networks or web servers are frequently used for evaluation of interaction sites while some researchers have employed evolutionary algorithms for prediction of the effective structure of antibodies. Accordingly, 14 methods based on the protein spatial structure, 28 researches based on the molecular amino-acide sequence (without usage of the spatial structure), and 18 antigen/antibody structure prediction techniques were reviewed. Results: We demonstrated that the accuracy of structure-based methods can be increased up to 80% while the acuracy of sequence-based methods was rarely better than 75%. Since the spatial structure of many antibodies is unknown, some researchers raised the accuracy (even to 96%) by only antibody sequences able to interact with some similar antigens in training neural networks. Therefore, we suggest this approach for structure prediction of monoclonal antibodies because of its adequate high accuracy. Conclusion: In this paper, after reviewing available methods for prediction of antibody-protein interaction sites, some suggestions were made for effective prediction of structure of monoclonal antibodies.

Keywords

• immunology
• monoclonal antibodies
• antibody-protein complexes
• artificial intelligence
• neural networks