Test-Time Structure-Space Search lifts antibody design success from 16% to 78%
anshulkundaje · x · 2026-10-09
The OpenDDE team released Test-Time Structure-Space Search, a new inference algorithm for designing and selecting antibody drug candidates.
Key result: on antibody–antigen targets where OpenDDE usually fails, top-1 success jumps from 16% to 78% with a frozen model — no retraining, no gradients through the network, just five correct contacts.
- With the same budget of 25 candidates per target, re-ranking with contacts only reaches 33%, while steering the sampling trajectory during denoising reaches 78% — roughly three quarters of the gain comes from changing the sampler's trajectory, not post-hoc selection.
- The idea: test-time compute isn't only for LLMs — instead of picking the best of many samples, search for better structures while the sampler is still denoising.
Backed by the open-source repo aurekaresearch/OpenDDE (505 stars, Apache-2.0), an all-atom biomolecular foundation model turning co-folding into a scalable engine for structure prediction, design, and optimization in drug discovery. It's a preview release; CLI flags and checkpoints may change without reproducibility guarantees.
More from Research
- Zero human labels: auto-generated soccer tracking dataset hits 62.5 HOTA, beating fine-tuned FairMOT — RexDouglass · 2026-10-09
- Why AI doesn't actually read words: from BPE subwords to byte-level models like BLT and H-Net — jbhuang0604 · 2026-10-09
- NVIDIA details HSTU recommender inference stack with up to 5.93x lower latency — PyTorch · 2026-10-09
- Preprint: LLMs store numbers as curves and helices, but compute comparisons differently — tweetsatpreet · 2026-10-09
- Best model was cheapest: open-weights model ran 669 clinical decisions for 1.7 cents — antoine_chaffin · 2026-10-09
- Ofir Press Points to ExcelBench as the Right Scale for Benchmarking Coding Agents — OfirPress · 2026-10-09