Optimization of Antibody Candidates to Mutated Antigens via a Paratope-Informed Pipeline

Antibody optimization is critical for improving binding affinity and stability against mutated antigens. However, state-of-the-art methods often overlook the binding mode between the antibody and the original antigen, limiting their effectiveness. Thus, we integrate the analysis of binding mode into the antibody optimization process. To be specific, we propose a computational pipeline that integrates paratope identification, antigen alignment, and residue-level optimization, leveraging a diffusion-based generative model (DiffAb). We validated the pipeline through experiments on antibody restoration and optimization. Key evaluation metrics, including Amino Acid Recovery Rate (AAR), Cα Root-Mean-Square Deviation (RMSD), Epitope Hit Ratio (EHR), and Improvement in Curvage Error Change (ICEC), guided the selection of candidates for molecular dynamics (MD) simulations. In antibody restoration experiment, our pipeline demonstrated competitive performance against baselines on CVB1-related antibody-antigen complexes. For antibody optimization, we successfully refined CVB5-binding antibodies to interact effectively with CVB1 and CVB3 antigens. MD simulation results further confirmed the stability of the optimized antibodies in binding mutated antigens. These findings highlight the pipeline's capability to adapt antibodies to mutated antigens, providing a robust and versatile tool for therapeutic antibody design.