Innovations in artificial intelligence (AI) have quickly revolutionized the field of structural biology making it possible from predicting protein folding and their three-dimensional structures up to creating de novo protein structures from scratch. By incorporating structure prediction models with generative diffusion models, AI-assisted approaches represent a paradigm shift in molecular engineering. With the help of these tools, targeted protein design is enabled aiming at specific protein-protein interactions to generate small proteins selectively binding to the desired epitope. This rapid process massively elevates the development of anti-cancer biologics like antibody-drug conjugates (ADCs), chimeric antigen receptor (CAR) T cells, and bispecific T cell engagers (TCEs), which rely on highly specific target recognition of tumor-associated antigens and minimal off-target interactions.

Despite these advances, several challenges remain, such as the binder design efficiency being variable and strongly dependent on the target structure and chosen epitope. Nectin-4 is a well-characterized cell adhesion molecule overexpressed in various solid cancer entities including urothelial carcinoma, breast, and lung cancer. Designing binders against this target proved to be particularly inefficient making Nectin-4 a rather challenging target compared to structurally similar surface proteins like PD-L1. Our goal is to improve de novo protein design against tumor antigens to accelerate therapeutic protein engineering and development of new targeted cancer agents.