OpenADMET runs community blind challenges to benchmark predictive models on realistic drug discovery datasets. These challenges create rigorous, transparent tests of performance while helping release valuable datasets and methods to the broader community.
A growing archive of community challenges built around realistic experimental datasets.
A blind challenge on cytochrome P450 inhibition across CYP3A4, CYP2C9, CYP2D6 and CYP1A2, built with Octant and the UCSF Fraser lab, covering direct inhibition (pIC50 regression) and time-dependent inhibition (classification) on Enamine DDS10 compounds and FDA-approved drugs.
A blind challenge on human PXR induction spanning both an activity prediction track and a structure prediction track, built on the largest, most consistent high-quality PXR induction dataset ever made publicly available.
A lead-optimization-style blind challenge based on real-world ADMET data from Expansion Therapeutics. Participants predicted nine ADMET endpoints using earlier-stage molecules to forecast late-stage compounds.
An earlier OpenADMET-associated community blind challenge focused on pan-coronavirus drug discovery data, bringing together participants to evaluate computational methods on realistic potency and structure tasks.