OpenADMET

OpenADMET is an open effort to build predictive models of ADMET properties and understand the mechanisms by which they arise. This means systematically characterizing the proteins and mechanisms that give rise to these properties (protein structures and scaled functional and genetic assays), understanding how small molecules interact with these mechanisms (through high-throughput nanoscale chemistry to explore/exploit chemical space), and integrative computational models (AI/ML ligand/structural, mechanistic, and physiological modeling).

We see our role as a guide to the community by developing open datasets and computational models. One way to ignite innovation in ADMET modeling is through community blind challenges. Blind challenges can provide accurate benchmarks of current performance and help us understand how much we have left to achieve. The paragon example is the CASP challenge that set up conditions for the “AlphaFold” breakthrough in protein structure prediction. For ADMET challenges, we plan on using both our generated data on anti-targets of broad interest and ADMET data donated from the community, as we did in our first challenge with the ASAP AViDD center.

OpenADMET is a nascent coalition of aligned efforts funded by different organizations. Our work currently involves personnel at OMSF, UCSF, Octant, and MSKCC. Our initial funding is through an ARPA-H grant: “AVOID-OME”. Since then, we’ve additionally been funded by the Gates Foundation & Schrodinger, to expand into toxicity and fundamental molecular properties, and by the Astera Institute, to expand our metabolism dataset coverage. We are hard at work on these efforts, and please stay tuned as we begin to launch some new competitions and ways to get involved over the next year.

Website: openadmet.org
Github: github.com/OpenADMET