Berlin’s Apheris and Ginkgo bring pharma giants together to train AI on 10,000 antibodies

Berlin-based Apheris and Ginkgo Datapoints have formed the Antibody Developability Consortium, joining forces with industry giants AbbVie, argenx, Lundbeck, and Takeda. This collaboration aims to create the most extensive, standardized antibody developability dataset ever assembled, containing 10,000 antibodies. By combining proprietary antibody sequences with Ginkgo's publicly sourced data, the consortium hopes to overcome the limitations of existing predictive models that suffer from small, fragmented, and inconsistent datasets.

Utilizing Apheris' federated data networks, members can train, benchmark, and refine AI models on the full dataset without exposing raw proprietary sequences to other participants. This allows pharmaceutical and biotech companies to develop models trained on the complete consortium dataset and apply them internally while retaining ownership of their proprietary sequences and assay data.

The consortium's scientific design, implementation, and high-throughput wet-lab characterisation will be conducted by Ginkgo Datapoints, while independent scientific oversight will be provided by Professor Charlotte Deane from Oxford and Professor Peter Tessier from Michigan. The initial dataset is expected to be delivered to members by early 2027, with plans to expand the dataset to include more complex antibody formats in the future.