A One-shot Framework for Directed Evolution of Antibodies

ICLR 2026 Conference Submission16802 Authors

Keywords: One-shot learning, matching, directed evolution, antibodies, structure-embeddings

TL;DR: AffinityEnhancer boosts antibody affinity from a single sequence, without structure or antigen data, by learning from matched high– vs. low-affinity pairs, capturing generalizable binding features and outperforming SOTA baselines.

Abstract: Improving antibody binding without extensive data or antibody-antigen structure information remains a critical challenge in therapeutic protein design. In this work, we propose, AffinityEnhancer, a framework to improve the affinity of an antibody in a one-shot setting. In the one‐shot setting, we start from a single lead sequence—never fine‐tuning on it or using its structure in complex with the antigen or epitope/paratope information—and seek variants that reliably boost affinity. During training, AffinityEnhancer utilizes pairs of related sequences with higher versus lower measured binding in a pan-antigen dataset comprising diverse “environments” (antigens) and a shared reconstruction module that learns to transform low‐affinity sequences into high‐affinity ones, effectively distilling the consistent, causal features that drive binding. By incorporating pretrained sequence–structure embeddings and an antibody-specific sequence decoder, our method enables robust generalization to entirely new antibody seeds. Across multiple unseen internal and public benchmarks, AffinityEnhancer outperforms existing structure‐conditioned and inpainting approaches—achieving substantial (in-silico) affinity gains in true one‐shot experiments without ever seeing antigen data.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Submission Number: 16802