Since the approval of the first monoclonal antibody, antibodies have become one of the most versatile and diagnostic agents due to their high specificity and affinity for a wide variety of targets. In the process of antibody development, the optimisation of antibody binding affinities and specificities can be a significant challenge While computational approaches can dramatically reduce the time and costs associated with affinity maturation, current methods are of limited accuracy.
We previously showed that using graph-based signatures to describe the physicochemical properties and geometry of the antibody-antigen structure could rapidly and accurately predict the effects of missense mutations on antigen binding affinity. In order to better guide rational antibody engineering, we have curated a new mutational database with 754 new mutations with experimental data. We then explored the combination of our graph-based signatures with structure-based signatures, energy-based functions and evolutionary conservation. The final model significantly outperformed available tools, achieving a Pearson's correlation of 0.76 on 10-fold cross-validation and 0.64 on non-redundant blind tests. Even when homology models were used, built on templates down to 25% sequence identity, the model still achieved a Pearson’s correlation of 0.72.
We have implemented our new approach as a user-friendly web-server that enables rapid evaluation of specific mutations, or comprehensive alanine or saturation mutagenesis. This in silico approach will play a crucial role in providing information about not only improving the affinity but also studying escape mutations of therapeutic antibodies. Users can freely use mCSM-ABv2 for rational antibody affinity optimisation at http://biosig.unimelb.edu.au/mcsm_ab2.