ProteinDPO Adopts DPO to Align Protein Models with Experimental Stability
bravo_abad · x · 2026-08-23
Talal Widatalla and coauthors introduce ProteinDPO, adapting Direct Preference Optimization (DPO) from LLMs to protein language models. This addresses the misalignment between the model's internal notion of a "good protein" and actual engineering properties like stability.
Key Methodology:
- Utilizes approximately 660,000 experimental stability measurements as preference data.
- The model learns by comparing more and less stable variants rather than fine-tuning solely on good examples, preserving general knowledge from pre-training.
Outcome: The model, fine-tuned from ESM-IF1, shows improved alignment with experimental preferences.
More from Research
- WEDGE paper generates training environments conditioned on navigation agent failures — _akhaliq · 2026-10-16
- RExBench: best coding agent completes only 33% of autonomous research extensions — najoungkim · 2026-10-16
- Google's CLASHarness benchmarks multi-agent conflicts in computer use, finds frontier VLMs struggle — google · 2026-10-16
- Tsinghua team proves vanilla flow-matching sampling can't exceed constant-depth expressive power — SQZ · 2026-10-16
- APEX-Harm benchmark: script-based injection attacks easily compromise agents in shared workspaces — KAUST · 2026-10-16
- Residual-stream forecasts predict agent violations before they happen, USC study shows — USC · 2026-10-16