Date:
Type:
Publication:
Author(s):
Antibodies have become indispensable in drug discovery and diagnostics, yet their development is rate limited by animal-based in vivo affinity maturation and monoclonal antibody isolation. However, computational approaches for antibody design necessitate innovation. For example, while protein structure predictions have radically improved, antibodies pose a continued challenge due to their rapid and flexible sequence evolution. In addition, free energy perturbation (FEP) for protein-protein systems requires sampling of complex configurational landscapes. Here, we introduce AQFEP for biologics, a scalable and robust absolute alchemical FEP methodology for predicting the affinity between macromolecular binding partners. The approach is versatile, as we benchmark its performance using both high-resolution crystal structures and AI-predicted models generated with our AQCofolding algorithm. As part of our methodology, we employ deep learning-driven side-chain refinement to optimize the binding site complementarity. This step ensures an accurate depiction of the molecular recognition environment and enhances the fidelity and predictive power of free energy calculations.
Through our methodological advancements in system preparation and increased alchemical sampling, AQFEP provides robust predictions of absolute binding affinities. For benchmarked antibody-antigen complexes, including wild-type and multi-mutant variants, our protocol achieves a Spearman correlation of 0.67 for crystallographic structures. We will present our performance for FEP challenges in biologics, including sampling convergence and handling flexible interfaces. By efficiently filtering out poor binder sequences from experimental methods to measure binding affinities, AQFEP shows potential to accelerate antibody design and significantly reduce reliance on animal models in monoclonal antibody development.
