Back

Structure-based design of antibody repertoires with drug-like properties

Tennenhouse, A.; Wilen, R.; Gaiduk, S.; Borni, S.; Yardeni, E. H.; Goldsmith, M.; Albeck, S.; Unger, T.; Diskin, R.; Fleishman, S. J.

2025-12-10 synthetic biology
10.64898/2025.12.10.693474 bioRxiv
Show abstract

Animal immunization is the prevalent strategy for discovering antibody therapeutics, but it is a lengthy and poorly controlled process. As an alternative, antibody repertoires can be synthesized by recombining human antibody genes and introducing random mutations in their complementarity-determining regions. Synthetic repertoires can deliver binders quickly and without associated animal-welfare concerns, but the resulting antibodies often fail to exhibit "drug-like" biophysical properties. We developed a principled structure- and energy-based strategy, called CADAbRe, for designing repertoires of human antibodies that are stable by design. Every antibody was programmed to ensure its stability while maximizing the structural diversity of the entire repertoire. We also developed a cost-effective strategy for repertoire assembly and, as a proof of concept, synthesized over 500 million unique antibodies based on hundreds of human antibody gene combinations and designed CDR H3 sequences. Structurally diverse binders against four unrelated antigens exhibited affinities, specificities, and drug-like properties seen in therapeutic leads. CADAbRe is the first programmable structure-based strategy for antibody-repertoire design, enabling design-test-learn cycles towards high-quality universal and customized repertoires. We envision that these will accelerate and rationalize basic and applied antibody discovery and optimization while addressing animal-welfare concerns. The repertoire will be available for academic research.

Matching journals

The top 3 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.