Deep Origin deploys AI-powered ADMET models across drug discovery programmes
Read Full Article on BioSpectrum Asia
BioSpectrum Asia covers the first pilot program from Deep Origin's ARPA-H-funded PREDICTS consortium, in which four drug developers are running 62 in silico ADMET models, 89% of which outperform literature benchmarks.
BioSpectrum Asia reports that Deep Origin and its PREDICTS consortium partners have launched a pilot program putting AI-driven drug safety prediction models to work inside four drug development organizations. The consortium is funded through ARPA-H’s CATALYST program, with an award of approximately $31.7 million made in 2025. The four pilot participants span distinct therapeutic areas: ImmVue (immunology), Sanford Burnham Prebys (oncology and immunology), Synko (antivirals), and SyzOnc (oncology).
The article describes ADMET-NOW, a suite of 77 machine learning models predicting absorption, distribution, metabolism, excretion, and toxicity properties, 62 of which are available to pilot participants. A foundation chemistry model called Togo lets the endpoint-specific models perform with limited data, often fewer than 1,000 data points per endpoint, and 89% of the deployed models outperform literature benchmarks, including on genotoxicity and drug-induced liver injury. Co-founder and Chief Business Officer Natalie Ma notes that lab results from the pilot will feed back into the system to expand model performance. Looking ahead, the consortium plans to build virtual human avatars for toxicology simulation, with mechanistic models of the liver, kidney, intestine, blood coagulation, and bone marrow.
For our own announcement, see Drug Developers Deploy In Silico Drug Safety Prediction Models From Deep Origin’s ARPA-H-Funded PREDICTS Consortium.
