Performance evaluation of artificial intelligence methods in nanoantibody design

Abstract

Nanobody is an antibody found in naturally missing light chains in camel families such as alpaca and monama, as well as cartilage fish such as sharks and rays, including two constant regions, a hinge region and a heavy chain variable region. With its small size, high stability, strong affinity, low cytotoxicity, strong penetration, and simple humanization, it is widely used in disease diagnosis, treatment and novel nanodrug design. Traditional nanobody acquisition methods for animal immunization or library screening. The traditional preparation methods have the disadvantages such as cumbersome process, poor specificity, difficult protein expression, and inability to target specific epitopes. Therefore, innovative strategies are needed to transform the sequence and structure of nanobodies and design new antibodies that are not available in nature. Nowadays, through the artificial intelligence method to deep learn the complete information of the target antigen, the variable region of the antibody and the complex internal and external physical effects of the antibody, which can effectively generate the 1D sequence and 3D structure of the antibody CDR region. Nanobodies designed by artificial intelligence have strong antigen targeting characteristics, expression ability and generalization ability, and have wide application prospects. Taking diffab and AlphaPanda as an example, this paper introduces the design method of AI antibody in detail from the aspects of model building and antibody design process, and evaluates the design performance of RMSD, Seqid and ddG. The results show that the RMSD values of CDR1 and CDR3 are greater than 2Å; CDR2 is less than 1.5Å, reaching atomic accuracy. Only the diffab designed CDR2 sequence showed good agreement, numerically over 30%.0.0067% of the CDR designed by diffab was less energetic than the natural antibody and 0.0233% of the CDR designed by AlphaPanda was lower than the natural antibody. The overall performance of the designed CDR is better in the above indicators. Based on the above data, this paper proposes improvement measures for the AI antibody design program, and puts forward new ideas and prospects for the future field of AI antibody design.