Assessing the state of the art in ADMET prediction: Lessons from the OpenADMET-ExpansionRx community blind challenge
This paper reports the results of a community blind challenge organized with Expansion Therapeutics, where participants predicted nine ADMET endpoints from a realistic dataset. The top performers were message‑passing neural networks and chemical foundation models, and the study highlighted the importance of proprietary industry data and the need to tailor predictive workflows via multitasking and task‑affinity grouping. Hosting the challenge on Hugging Face proved pivotal for community engagement and visibility, underscoring the value of robust, open‑source infrastructure for advancing ADMET …
This paper reports the results of a community blind challenge organized with Expansion Therapeutics, where participants predicted nine ADMET endpoints from a realistic dataset. The top performers were message‑passing neural networks and chemical foundation models, and the study highlighted the importance of proprietary industry data and the need to tailor predictive workflows via multitasking and task‑affinity grouping. Hosting the challenge on Hugging Face proved pivotal for community engagement and visibility, underscoring the value of robust, open‑source infrastructure for advancing ADMET …
The findings demonstrate how Swiss AI initiatives can leverage open‑source blind challenge platforms, advanced neural architectures, and industry‑specific data strategies to improve ADMET predictions in drug discovery.
Original source record. Open the original record to verify the underlying announcement.