Apheris and Ginkgo launch antibody AI consortium with four pharma companies
The AMW Read
A four-company pharma consortium with a planned 10,000-antibody dataset meaningfully expands the data available for developability modeling, though performance gains remain unreported.
Apheris and Ginkgo launch antibody AI consortium with four pharma companies
Berlin-based Apheris and Ginkgo Datapoints have launched the Antibody Developability Consortium with AbbVie, argenx, Lundbeck, and Takeda as founding members. Each pharma company will contribute proprietary antibody sequences, while Ginkgo will add publicly available sequences as needed to reach a dataset of 10,000 antibodies. Ginkgo will oversee antibody production and laboratory characterization, then train a developability model within Apheris's secure environment. Members can train and benchmark models on the combined dataset without exposing their raw proprietary sequences to one another.
The collaboration targets a data bottleneck in AI-assisted drug development. An antibody can show promise as a treatment yet prove difficult to manufacture or formulate. Predictive models for those risks have been constrained by fragmented datasets and limited sequence diversity, including within individual companies. The consortium pairs standardized laboratory measurements with privacy-preserving access to data held across several drugmakers. That could give participating teams a broader basis for selecting candidates earlier, while letting each company retain ownership of the sequences and assay data it contributes.
For drug developers, the practical test will be whether models trained on the shared dataset predict developability more reliably on their own candidates than models trained on internal data alone. Ginkgo and Apheris describe the consortium's design and intended dataset, but the article reports no measured improvement or clinical outcome. Investors evaluating similar collaborations should look for evidence of performance across members' distinct antibody portfolios, alongside proof that data access remains workable as the consortium grows.