Ligo's Artios designs functional de novo enzymes with all-atom diffusion, 5x better on AME benchmark
rishabh16_ · x · 2026-10-21
Ligo Biosciences published Artios, a two-stage de novo enzyme design system, and says it will fully open-source the code including for commercial use.
- ArtiosBackbone: an all-atom diffusion model that generates a protein backbone plus ligand around an atomic active-site motif and chooses catalytic residue positions.
- ArtiosPotts: a ligand-aware Potts-model sequence designer sampling from an explicit energy function, with new terms addable at sampling time without retraining.
- State-of-the-art on the Atomic Motif Enzyme (AME) benchmark; when the pocket must retain its ligand, it produces 5x as many passing designs as the strongest prior method. Designed sequences are more expressible than LigandMPNN's.
- In the lab, Artios produced multiple functional new-to-nature luciferases, esterases and cutinases, screening no more than a single 96-well plate per campaign.
Related event: Open-Source Enzyme Design System Artios Targets Protein Design Frontier(3 posts)→
More from AGI Musings
- "Open source values are dying": maintainer laments eroding gratitude in the AI era — facontidavide · 2026-10-22
- Harvard physicist says AI could solve fundamental physics in 5 years as accuracy jumps 2% to 35% — AryHHAry · 2026-10-22
- Defenders' AI, not attackers, found the vulns — fueling AI offense-defense debate — cephaloform · 2026-10-22
- 'Agent + cybersecurity will be HUGE': devs debate returning to security careers — max_paperclips · 2026-10-22
- Demis Hassabis: today's AI will look like nothing compared to what's coming in 1–2 years — VraserX · 2026-10-22
- AI Security Enters a Machine-Speed Arms Race, Human Patch Cycles Are Dead — srimisra · 2026-10-22