An iterative strategy to design 4-1BB agonist nanobodies de novo with generative AI models.

The 4-1BB receptor, a key member of the tumor necrosis factor receptor (TNFR) family, represents a highly promising target for cancer immunotherapy. In this study, we developed a novel in silico pipeline to design VHH domain antibodies targeting 4-1BB, leveraging knowledge-based amino acid distributions to generate optimized complementarity-determining region (CDR) sequences. Our computational approach progressively refined nanobody binding properties, yielding designs with binding scores comparable to or exceeding those of an established reference nanobody. From an initial set of 80 top-ranked de novo sequences, 65 were successfully assembled, with 35 validated by sequencing. Although this screening round did not yield a high-affinity binder in vitro, the results provide critical insights into the relationship between initial design parameters and successful genetic assembly. These findings highlight the potential of our pipeline while identifying key areas for further refinement, particularly in optimizing deep-learning models for antibody development. This work advances the broader effort to harness computational design for high-precision therapeutic antibody discovery.