Five pharma giants federally train OpenFold3 on 20,000 private structures, +11 pts
jiqizhixin · x · 2026-10-16
A Nature paper (Sept 14, 2026) details how AbbVie, Astex, Bristol Myers Squibb, Johnson & Johnson and Takeda federally trained OpenFold3 via Apheris:
- Data: 20,000+ proprietary protein–small-molecule structures, roughly tripling drug-relevant training data; raw data never leaves each company, only model parameters are aggregated
- Results: 11 pts accuracy gain over the strongest public-data model, beating any single-company model; under 10 weeks from first run to SOTA
A landmark real-world case of "move the model, not the data" for AI-driven drug discovery.
More from Venture
- Box CEO Aaron Levie: agent wallets finally make micropayments a good business model — kleffew94 · 2026-10-16
- Surplus incubator offers $100k SAFEs for AI-era for-profit public-good startups in SF — NathanpmYoung · 2026-10-16
- fomo.family creator rewards hit $102,985 for top user just 6 weeks after launch — benaratame · 2026-10-16
- Cal AI banked $30M in two years on influencer marketing, sold to MyFitness — TechCrunch AI · 2026-10-16
- VC meme: today's $50M calendar-app rounds vs 1600s nutmeg fleet bets — ctjlewis · 2026-10-16
- Mark Cuban's "AI is book smart, not street smart" dubbed the best pitch for RL environment startups — joecole · 2026-10-16