Four pharma companies — AbbVie, argenx, Lundbeck, and Takeda — have signed on as founding members of the Antibody Developability Consortium, a collaboration organized by Ginkgo Datapoints and Apheris that aims to assemble what the partners claim will be the field's largest standardized antibody developability dataset, with a target of 10,000 antibodies.

The consortium remains open to additional pharma and biotech companies. Each founding member will contribute proprietary antibody sequencing data through Ginkgo Datapoints, an offering of Ginkgo Bioworks, with remaining capacity filled from publicly available sources.

The business case is straightforward: antibody developability — the set of biophysical properties that determine whether a candidate can be manufactured, formulated, and advanced into a clinical product — is a major source of late-stage attrition. Predicting manufacturability barriers earlier could change candidate selection decisions and cut wasted development spend.

Federated model, private sequences

The architecture reflects a familiar data-governance constraint: pharma companies will not share proprietary sequences. Under the consortium design, Ginkgo Datapoints leads scientific design and execution — sequence selection, antibody production, and high-throughput wet-lab characterization across core developability endpoints. Ginkgo is also training a foundation antibody developability model on the dataset inside Apheris' secure environment.

Apheris' federated infrastructure then delivers that foundation model into each member's own environment, where members can fine-tune it on proprietary data. Members can also train, benchmark, and refine their internal models in Apheris' environment using the full consortium dataset plus their own sequence data, without exposing proprietary information to other participants.

"This consortium represents an important step forward in building predictive models for antibody developability by creating datasets that are designed for machine learning, addressing limitations associated with convenience datasets," said Athena Hadjixenofontos, PhD, director of data science, head of AI in biotherapeutics and genetic medicine at AbbVie. "Federated infrastructure enables participants to contribute data while keeping proprietary sequences private. These capabilities could meaningfully accelerate antibody discovery and help advance new medicines for patients."

The reference to "convenience datasets" is a pointed one: much of the public developability literature rests on small, heterogeneous datasets assembled for other purposes, which limits how well models trained on them generalize.

Independent oversight and timelines

The consortium has appointed two independent scientific advisors: Charlotte Deane, PhD, professor of structural bioinformatics at the University of Oxford, and Peter Tessier, PhD, professor of pharmaceutical sciences and chemical engineering at the University of Michigan. The announcement did not specify the scope of their oversight role.

For Lundbeck, the motivation is therapeutic-area specific. "In complex therapeutic areas such as CNS" the "ability to select well behaved candidates with superior developability properties is essential," said Allan Jensen, PhD, vice president, biotherapeutic discovery at Lundbeck. "By bringing together diverse antibody datasets, this collaboration has the potential to strengthen predictive approaches."

Members should get access to the initial dataset by early 2027. The partners also plan to expand the dataset over time to include more complex antibody formats, enabling new drug classes and additional properties relevant to predicting which candidates will succeed or fail.

"Pooling standardized developability data across the industry can create stronger predictive models than any one company could build alone," said Yves Fomekong Nanfack, PhD, head of AI/ML research at Takeda. "As we advance Takeda Research's ambition to become an AI-native discovery organization, this capability can help identify promising antibody candidates earlier, inform better development decisions, and bring new therapies to patients faster."

What to watch

Several claims warrant scrutiny as the consortium matures. The 10,000-antibody target and the early-2027 access date are plans, not measured results. The partners have not disclosed membership fees, the specific developability endpoints to be characterized, or how much of the dataset will come from public sources versus proprietary contributions. The value of the foundation model will depend on whether the standardized wet-lab characterization is consistent enough across batches to support reliable machine learning — a known challenge in high-throughput biophysical assays.

For R&D managers weighing similar precompetitive data-sharing arrangements, the structure here — federated fine-tuning on a shared foundation model, with independent academic oversight — offers a template worth tracking. The first test will come in early 2027, when members begin working with the initial dataset and can benchmark it against their internal models.