Nature: Large blind benchmark of AI-designed antibodies shows promise and ceiling
ferruz_noelia · x · 2026-08-22
- Scale: First large blind benchmark involving 29 independent labs testing 511 AI-designed antibody binders, designed with zero knowledge of real target structures.
- Results: A handful of AI-designed antibodies achieved sub-100pM affinity in wet-lab tests, rivaling approved therapeutics.
- Limitations: Most methods failed to generalize past the targets they were tuned on, revealing a ceiling in current generalization capabilities.
- Significance: Published in Nature Biotechnology, this prospective benchmark anchored to experimental data highlights both the promise and the current limits of computational antibody discovery.
More from Research
- Interpretability paper incoming: in-depth look at testing explanations on modern LLMs — burny_tech · 2026-10-16
- Better-learning models tend to be more interpretable, argues Nick Cammarata — burny_tech · 2026-10-16
- Cornell and Amazon revisit the 16x pretraining compute gap: diffusion LMs need only 3-8x — LucaAmb · 2026-10-16
- AI trained on the fraction of human intelligence that fit the page — r0ck3t23 · 2026-10-16
- Quotient Recurrent Reasoners hit 97.9% on Sudoku-Extreme with 78% fewer loops — burny_tech · 2026-10-16
- Mathematicians publish explainer for the Egyptian fraction proof in OpenAI's proofs collection — burny_tech · 2026-10-16