Private pharma data turbocharges AI protein-folding models
A consortium of drug companies trained the OpenFold3 AI protein-folding model on over 20,000 proprietary protein structures, including protein–ligand complexes, achieving markedly better predictions than models trained on public data alone; the approach demonstrates the value of pooled pharma data for drug discovery, though the AISB results are not yet peer-reviewed and the model isn't publicly released.