Four Pharma Giants Begin Federated Training on Binding Affinity Models

What changed

The AI Structural Biology Network, hosted by Apheris, launched AISB Bind on 22 September 2026. AbbVie, AstraZeneca, Bristol Myers Squibb, and Johnson & Johnson will collaboratively train a model to predict how small molecules bind protein targets well enough to guide virtual screening and lead optimization.

Training uses federated learning: each partner keeps proprietary data inside its own secure environment while Apheris orchestrates rounds on Amazon Web Services infrastructure. Partners receive the resulting model to run and further tune locally. Federated training begins in September 2026 with final models expected in early 2027.

Why it matters

Public structure models often fail on proprietary chemotypes seen only inside pharma vaults. A consortium model trained across four major pipelines could raise hit quality on industrial data without exposing structures, shifting early discovery from single company fine tunes to shared baselines with local adaptation.

The network positions models as modular, agent callable blocks inside R and D workflows, which aligns with growing agentic chemistry stacks in large pharma IT roadmaps.

Who is affected

Medicinal chemistry leaders at mid and large pharma; computational biology vendors competing with consortium baselines; CROs selling virtual screening services that must benchmark against AISB Bind outputs once released.

What to do next

If you compete with these four companies in small molecule discovery, stress test your in house binding models against public baselines now, before AISB Bind sets a new federated reference in early 2027.

What to watch

Publication of validation metrics on held out partner splits and whether additional AISB Network members join beyond the four launch partners.

Sources

1. Primary. AISB Network via Business Wire, AISB Network Launches AISB Bind (22 September 2026). Establishes partners, federated design, and timeline.