Evo 2 genome design: 4-8 natural genomes in prompt lifts pass rate from 13% to 52%
garykbrixi · x · 2026-10-29
The author reports that for genome design with Evo 2, putting four to eight natural ΦX174-like genomes in the prompt beat prior fine-tuning on thousands of sequences: 52% of designs pass correctness filters versus 13% for the fine-tuned model, and a library of 64 novel genomes takes 2.1 GPU-hours instead of roughly 200.
Changing only the prompt extends the approach to the RNA phage MS2 and the filamentous phage M13.
In a follow-up thread the author adds antibody results: with 8 de novo binders in the prompt and no structural information, in-context learning produced the most passing designs per GPU-hour of any method tested, and was the only method to find a unique hit for all 7 antigens, with designs contacting the intended epitope.
More from Research
- NVIDIA's AgentBaton cuts agent API costs 65.9% with a single model handoff — dair_ai · 2026-10-30
- Stanford releases RT-J, a System 1 model predicting structured data in one forward pass — jure · 2026-10-30
- Generalist AI ships GEN-1.5 and challenges the community to SLAM its wrist-cam data — E0M · 2026-10-30
- Signal65 Launches PINNACLE Coding Agent Benchmark: One Model Aces 97% of Samples, Fails All 30 Jobs — ryanshrout · 2026-10-30
- TEM: Transformer Energy-Based Models Claim to Outscale Diffusion Without Guidance — du_yilun · 2026-10-30
- VisualForce: robots infer grip force from soft gripper deformation, no force sensors needed — du_yilun · 2026-10-30