De novo protein binder design

Establish a validated methodology for de novo protein binder design that directly optimizes binding affinity, including identifying reliable reward functions (e.g., alternatives to interfacial predicted aligned error, iPAE) suitable for guiding generative search and evaluation.

Background

The paper demonstrates affinity maturation experiments using beam search guided by rewards computed via ProteinMPNN and AlphaFold-Multimer, with iPAE as the primary metric. While these experiments show potential for improving affinity starting from weak binders, the authors caution that iPAE is not a perfect reward and that binder design, particularly de novo design, remains unresolved.

They emphasize the need for more reliable reward functions and protocols that correlate well with binding affinity and functional outcomes, noting that full de novo binder design requires sidechain awareness and robust validation. This situates binder design as an explicit open problem in the broader protein generation and evaluation pipeline.