Building fine-tuning capabilities for co-folding models in pharmaceutical research

Five pharma companies fine-tuned OpenFold3 Preview 2 across 20,167 private structures, none of which left their environments. On 1,056 held-out structures, high-quality interface predictions rose from 35.6% to 52.1% and correct ligand poses from 28.9% to 46.8%, ahead of every public model. A first for federated drug discovery.

We rebuilt the SandboxAQ affinity head inside ApherisFold and fine-tuned it on 49 JAK2 macrocycles. Architectural changes alone lifted Spearman ranking from 0.418 to 0.60; fine-tuning then pushed validation Pearson from 0.23 to 0.76, mostly by correcting how the model handled inactives.

Delivering a state-of-the-art federated co-folding model across five pharma companies in record time