Can LigandMPNN be used for Antibody Design
I watched a webinar by iGEM Sheffield on protein (re)design with machine learning, and wanted to know if this could be applied to antibody design. Antibodies are becoming more common drugs, being highly targeted and specific, so if they could be computationally designed for a target of interest, then it could be cheaper to use them clinically.
Rangel et al.1 demonstrate the use of LigandMPNN for redesigning antibody complementarity-determining regions (CDRs), which are then transplanted into a standard antibody scaffold. Although this isn't a redesign of the complete antibody, the CDRs are the region of the antibody with the greatest variability, and so they are the most important regions of the protein for binding to the target epitope.
Fischman and Ofran2 say that screening for a target antibody is a complex and laborious process, which frequently yields antibodies with poor biological activity (even if the binding interaction is favourable). This de novo design of the antibody can identify the sequences that facilitate the strong binding interaction with the target antigen, and the desired biological activity in the remainder of the antibody can be engineered through transplantation of the binding domains. Computational design of antibodies could be used to design antibodies targeting the desired epitope, with the intended biological activity on a target protein. Being able to do computational design could result in cost savings and faster research, lowering the cost of antibody-based drugs, and speeding up bringing these biological drugs to clinical use.
X-ray crystal structures can also be used to develop more effective antibodies, improving binding interactions with target antigen.3 Using AbDesign, which combines RosettaFold with many combinations of antibody sequences, new structures can be created and evaluated for binding efficiency with the target. These new structures are created from combinations of sequences already identified. Although this method does not develop new sequences, it is still possible to generate a huge diversity of antibody structures to evaluate against the target epitope.45
References