Poster Presentation The 47th Lorne Conference on Protein Structure and Function 2022

Rational antibody affinity maturation using graph-based signature (#139)

Yoochan Myung 1 2 , Douglas Pires 2 3 , David Ascher 1 2
  1. Biochemistry and Pharmacology, Bio21 Institute, University of Melbourne, Melbourne, Victoria, Australia
  2. Structural Computational Biology and Clinical Informatics, Baker Heart and Diabetes Institute, Baker, Melbourne, VIC, Australia
  3. School of Computing and Information Systems, University of Melbourne, Computing and Information Systems, Melbourne, Victoria, Australia

Antibodies are essential biomolecules of our immune system. Their advantages over small-molecule and protein-based drugs as therapeutic options are their high binding specificity and binding affinity towards diverse antigens. While those benefits facilitated the use of antibodies as therapeutic and diagnostic agents, the process of affinity maturation via antibody engineering remains laborious and time-consuming.

With the emergence of  machine learning algorithms and the increase of available antibody databases, many in silico tools have been developed to predict the binding affinity of antibody-antigen complexes and the consequences of mutations, however, most available approaches are limited in performance, scalability and ease of use.

Here, we show a rational antibody engineering platform that can accurately predict the binding affinity (∆G, CSM-AB) and the binding affinity changes upon mutations (∆∆G, mCSM-AB2 and mmCSM-AB) of antibody-antigen complexes. In terms of predicting the binding affinity (∆G), CSM-AB outperformed available tools also in ranking antibody-antigen poses of docking benchmark sets (DOCKGROUND and ZDOCK). Moreover, we have demonstrated the accuracy of our methods (mCSM-AB2 and mmCSM-AB) on single- and multiple-point mutations. As for the prediction of antibody-antigen binding affinity changes upon single-point mutation, mCSM-AB2 showed better performance than available tools achieving a Pearson's correlation of 0.76 (RMSE = 1.68 Kcal/mol) and 0.77 (RMSE = 1.66 Kcal/mol) on 10-fold cross-validation and low-redundancy blind-test sets, respectively. For multiple mutation (2 to 14 mutations per case) predictions, mmCSM-AB showed the best performance overall across cross-validation and independent blind-test sets achieving Pearson's correlation of up to 0.95. 


We have implemented our tools as web-servers that enable rapid and deep scanning of antibody-antigen binding and systematic exploration of all possible combinations of single or a set of double and triple mutations across antibody-antigen interface residues. Our approaches will help to guide rational antibody engineering by analysing the landscape of antibody-antigen interfaces (CSM-AB) and the effect of introducing mutations (mCSM-AB2 and mmCSM-AB). These user-friendly web-servers are freely available at http://biosig.unimelb.edu.au/ab engin.

  1. Yoochan Myung, Douglas E V Pires, and David B Ascher. (2021). “CSM-AB: graph-based antibody-antigen binding affinity prediction and docking scoring function.” Bioinformatics.
  2. Yoochan Myung, Douglas E V Pires, and David B Ascher. (2020). “mmCSM-AB: Guiding Rational Antibody Engineering through Multiple Point Mutations.” Nucleic Acids Research.
  3. Yoochan Myung, Rodrigues, C. H. M., Ascher, D. B., & Pires, D. E. V. (2019). mCSM-AB2: Guiding Rational Antibody Design Using Graph-Based Signatures. Bioinformatics.