Radial awarded OpenADMET a grant to develop tools for blind evaluation of AI models in drug discovery

Radial awarded OpenADMET a grant to develop tools for blind evaluation of AI models in drug discovery

On August 24, OpenADMET announced that it had received a grant from Radial, the life sciences division of the Astera Institute. OpenADMET runs open evaluations of AI models that predict the properties of small molecules used to develop drugs. The grant will fund code, protocols, and public records for future blind evaluations.

Models used in drug discovery are often evaluated against previously published laboratory measurements. If these data were included in the training set or contain systematic biases, a high score provides little evidence that the model will accurately predict the outcome of the next experiment.

In a blind evaluation, the organizer keeps the results of new laboratory measurements confidential. Teams submit their predictions in advance, after which the system compares them with the experimental results. This approach tests how accurately a model predicts the properties of new compounds.

OpenADMET is already running the CYP Challenge. Participants predict how strongly 750 compounds inhibit the activity of four CYP enzymes, which are proteins involved in drug metabolism. The current leaderboard scores predictions against half of the test set. After the deadline on November 3, 2026, the evaluation will cover the complete test set.

Each evaluation requires a system for distributing data, accepting predictions, validating file formats, and calculating the metrics used to compare predictive accuracy. The grant will fund open-source code for these operations, guidance for organizers, and public records of the results.

OpenADMET plans to provide laboratories and consortia with ready-to-use code and protocols for their next blind evaluation. The results will be stored in a public, verifiable record.