Scalable Antigen-Antibody Binding Affinity Landscape: A Case Study with ENHERTU

ABSTRACT Optimization of binding affinities for antibody-drug conjugates (ADCs) is inextricably linked to their therapeutic efficacy and specificity, where the majority of ADCs are engineered to achieve equilibrium dissociation constants (K d values) in the range of 10 −9 to 10 −10 M. Yet, there is a paucity of published data delineating the optimal binding affinity or its range that ensures improved therapeutic outcomes for ADCs. This study addresses this issue by integrating structural biophysics within a scalable in silico workflow to generate antigen-antibody binding affinity landscapes, with a focus on Trastuzumab, a monoclonal antibody employed in the treatment of HER2-positive breast cancer. By leveraging high-throughput computational techniques, including homology structural modeling and structural biophysics-based K d calculations, this research puts forward a set of high-accuracy structural and intermolecular binding affinity data for Her2-Trastuzumab-Pertuzumab (PDB entry 6OGE). Beyond the design of Her2-targeting ADCs with enhanced efficacy and specificity, this scalable antigen-antibody binding affinity landscape also offers a technically feasible workflow for the high-throughput generation of synthetic structural and biophysical data with reasonable accuracy. Overall, in combination with artificial intelligence (e.g., deep learning) algorithms, this synthetic data approach aims to catalyze a paradigm shift in the discovery and design of antibodies and ADCs with improved efficacy and specificity. SIGNIFICANCE With Trastuzumab as an example, this study presents a scalable computational biophysical generation of antigen-antibody binding affinity landscapes, serving two purposes: design of Her2-targeting ADCs with enhanced efficacy and specificity and continued accumulation of synthetic structural biophysics data for the development of useful AI-based drug discovery and design model in future. This scalable approach is broadly applicable to databases such as Protein Data Bank.