Autonomous AI agent achieves closed-loop drug discovery, yielding 3 sub-nanomolar TREM2 binders

muni bio (Manifold Bio) partnered with Adaptyv Bio to run a closed-loop autonomous drug-discovery pipeline powered by NVIDIA's Proteina-Complexa framework: an autonomous research agent designed proteins, which were then validated in Adaptyv's automated wet lab, ultimately producing 3 sub-nanomolar binders against the target TREM2 despite lab cycles taking days to weeks.

Confirmed

• The agent designed proteins using NVIDIA's Proteina-Complexa model, exploring nearly 14,000 candidate designs
• In Adaptyv's wet-lab validation, 9 of 10 tested designs showed strong binding signals, with 3 reaching sub-nanomolar affinity
• The 3 binders target TREM2 and outperform prior leaderboard and hackathon-winning results
• The wet-lab phase took days to weeks

Why it matters

• The case demonstrates an agent stitching computational design and automated experimentation into a closed loop, achieving end-to-end autonomous drug discovery rather than stopping at in-silico prediction
• Multiple reposts emphasized that this validates the feasibility of autonomous AI research agents in protein design, offering a template for closed-loop studies on more targets

2026-08-19 ~ 2026-08-19 · 6 related posts

• muni AI Agent Discovers Three Sub-Nanomolar Protein Binders — AllThingsApx · 2026-08-19
• NVIDIA's Proteina-Complexa enables agents to discover sub-nanomolar protein binders — AllThingsApx · 2026-08-19
• muni_bio agent explores 14,000 protein designs using NVIDIA tools — AllThingsApx · 2026-08-19
• AI Agent Closes Design-to-Test Loop in Drug Discovery, Generating 3 Sub-nanomolar Binders — AllThingsApx · 2026-08-19
• NVIDIA Autoresearch Agent Generates Sub-nanomolar Protein Binders — AllThingsApx · 2026-08-19
• Agents enter wet lab: 3 sub-nanomolar protein binders designed autonomously — anshulkundaje · 2026-08-19