• Why AI co-folding is limited by data, not algorithms: the proprietary protein–ligand structures needed to train accurate models sit inside individual pharma companies, so no single organization has enough; federated learning solves this by training across companies without any data ever leaving its owner's environment.
• Proof from the largest pharma-to-pharma federated network: results from the AI Structural Biology (AISB) network show federated co-folding models outperform both public and single-company models, succeeding where public models fail, with data diversity as the deciding factor.
• From one network to an industry capability: how the approach extends to the broader ecosystem and across modalities and what it takes to put these models to work in drug-discovery programs (focusing across uses cases like structure, binding affinity, ADMET and antibody developability).
Robin Röhm, Chief Executive Officer & Co-Founder, Apheris
