2025 Antibody Developability AI Competition Results
The 2025 AbDev Competition with prizes by Ginkgo and GENEWIZ has ended!
An open benchmark event for antibody developability—driving innovation through shared data, transparent evaluation, and community engagement.
The Antibody Developability Prediction Competition is a first-of-its-kind benchmarking challenge.
Hosted by Ginkgo Bioworks' Datapoints in collaboration with HuggingFace.
Designed to advance the state of AI-driven biologics, the competition invites participants to test and refine predictive models for key developability traits—including thermal stability, self-association, and polyspecificity—using a high-quality, experimentally-validated dataset.
Drawing inspiration from the impact of open benchmarking efforts in other fields, this challenge brings together academic teams, startups, and industry experts to push the boundaries of antibody engineering. It's an opportunity to contribute to a shared scientific resource, build community, and help define the standards that will guide the next generation of therapeutic design.
The competition has concluded, and we are excited to announce the winners! In total 252 teams registered with 115 submissions for the held-out test set! We are thrilled to see the community's engagement and ingenuity using the high-quality GDPa1 dataset.
We received submissions from over 26 countries, 33 companies, and 38 universities! Winners differed by assay. Top Spearman's ρ values on the test set reached 0.708 (hydrophobicity) and 0.392 (thermostability), while for extrinsic, interaction-dependent properties ρ were lower, including 0.356 (polyreactivity), 0.337 (self-association), and 0.310 (titer).
Meet the teams whose top-performing models on the private held-out test set claimed a share of the $60k in prizes from Ginkgo Datapoints and GENEWIZ from Azenta Life Sciences!
Polyreactivity
Nimra Asi from Boehringer Ingelheim through Data Science Talent LLC
Self association and the Open Source Prize
Jiang Zhu from X UNFOLD
Thermostability
Aviral Singh, Daksh Sammi, and Rushikesh Jadhav from Microcrispr Pvt. Ltd.
Titer
Antonín Dušek as a hobbyist from UCT-Prague
Hydrophobicity
Anonymous
To check how your team did, check out the Competition Page on Hugging Face.
Significance
This competition is focused on predicting biophysical properties of antibodies that are key to the drug-like properties that make proteins into medicines.
Competition Details
Objective
Submit models that predict one or more of these biophysical properties for a private test set of 80 antibodies.
The best models on this test set will win prizes.
Winners will be announced in November 2025.
Developability Properties
💧 Hydrophobicity (HIC)
🎯 Polyreactivity (CHO)
🧲 Self association (AC-SINS at pH 7.4)
🔥 Thermostability (Tm2)
🧪 Titer
Recognition
There will be prizes for the best predictors in each of the 5 categories, as assessed on the private heldout set.
Submissions may be anonymous, but participants must identify themselves to be eligible for prizes.
There will also be an "open-source" prize for the top entry that is trained on only the 246 antibodies and publishes reproducible code.
Interested in Future Competitions?
Sign up to receive updates about future AI competitions and research from Ginkgo Datapoints
Stay Updated
Be the first to know about new competitions, datasets, and research opportunities.