Pharma consortium trains OpenFold3 on 20,000 proprietary structures, beats Boltz-2 with 50% accuracy

TheMoonMidas · x · 2026-09-14

A Nature news report describes a pharma consortium (AbbVie, Astex, and others) fine-tuning OpenFold3, an open-source replication of AlphaFold 3, on more than 20,000 proprietary protein structures from their internal databases.

• The resulting model achieved high-accuracy predictions on 50% of test cases, vs 40% for Boltz-2 and 33% for OpenFold.
• The edge comes from private data: protein–drug interaction structures locked in company vaults that public datasets lack.
• Findings show pharma's secret data markedly outperforms models trained on public data alone or on individual siloed datasets.

Related event: Five Pharma Firms Federated-Tune OpenFold3 on 20,167 Private Structures, Beating Public-Data Models(5 posts)→

Original post →

More from Research

• MIT, Stanford, Harvard and CMU researchers build Social Simulation Arena to benchmark simulators prospectively — _Hao_Zhu · 2026-09-15
• Post-training shift makes model distillation nearly impossible to detect, researcher argues — maksym_andr · 2026-09-15
• Insilico's AI-designed rentosertib shows biological age reversal across six proteomic clocks in Phase IIa — thione · 2026-09-15
• Anthropic's Claude formalizes Fermat's Last Theorem in Lean, largely autonomously in 11 days — thione · 2026-09-15
• DeepMind's AlphaGenome Atlas maps predicted effects of all 9B single-letter DNA variants, free — thione · 2026-09-15
• OpenAI says internal system solved 90-year-old Navier–Stokes Millennium Problem with a Lean proof — thione · 2026-09-15

Research channel →