Welcome to our quarterly newsletter, which outlines our progress, goals, and priorities. OpenADMET supports the drug discovery ecosystem with high-fidelity data, structural insights, and predictive models to identify and mitigate ADMET liabilities.

During a highly productive third quarter, the OpenADMET team secured additional funding to sustain our momentum over the next few years. While we continued to focus on metabolism, we expanded our scope to include absorption, distribution, and toxicity. Global participation in our blind challenges remains strong; we launched a new challenge focused on cytochrome P450 (CYP) inhibition, time-dependent inhibition, and structure prediction. In addition, we open-sourced our blind challenge infrastructure, which currently powers an OpenBind team challenge and will support another upcoming blind challenge in early 2027.

Blind Challenges

Our prospective community blind challenges continue to serve as a crucial benchmark for evaluating predictive models against unpublished, real-world experimental data.

• PXR Challenge Conclusion: Successfully wrapped up the PXR blind challenge with 370 participating groups from industry and academia. Participants brought innovative approaches to activity and structure prediction that we, as a community, can learn from.
• CYP Challenge Launch: Launched the CYP blind challenge backed by 90,000 new experimental data points covering key cytochrome P450 isoforms: CYP1A2, CYP2D6, CYP2C9, and CYP3A4. We have seen strong participation so far!
• Structural Context: Added 20+ new CYP3A4 Cryo-EM structures to power structure-based modeling during the challenge.
• Infrastructure Open-Sourced: Open-sourced our production-ready blind challenge platform infrastructure, enabling any group to deploy molecular prediction benchmarks in hours.
• Best Practices Manuscript: Released a comprehensive manuscript detailing best practices, evaluation metrics, and statistical methodologies established across our blind challenge series.

New Funding

We are deeply grateful for the new and expanded support from our funding partners, which allows us to broaden our scientific scope and scale our open infrastructure.

• OpenAI Foundation: Funding directed toward blood-brain barrier (BBB) penetration and transporter mapping.
• Gates Foundation: Grant support to extend our data generation and predictive modeling efforts into Toxicity.
• ARPA-H: Phase II funding to deepen our characterization of Metabolism across the Avoid-ome.
• Astera/Radial: Strategic grant funding dedicated to expanding and sustaining our blind challenge infrastructure.

New Compounds & Data

To build robust machine learning models, we continue generating large, standardized datasets that cover diverse chemical space.

• Compound Library Expansion: Funding from Enamine enabled the acquisition of another 10,000 compounds for physical assays.
• ReFRAME Screen: Screened the 12,362-compound Calibr-Skaggs ReFRAME library for PXR induction to identify new chemical matter that enables us to further understand this challenging target. 
• Deep Mutational Scanning: Generated high-throughput Deep Mutational Scanning (DMS) datasets across four CYP isoforms to assess genetic variability.
• Structural Ground Truth: Solved new high-resolution structures for CYP3A4 and CYP2D6.
• HT Chemistry Hit Expansion: Executed direct-to-biology high-throughput chemistry hit expansion for both PXR and CYP3A4.

Software & Open Models

In addition to releasing datasets, we are developing and supporting a set of open-source ADMET models. Many of these models are direct results of ideas developed by participants in the blind challenge. 

• Participant-Informed Models: Deployed new open-access machine learning models incorporating top-performing architectures, feature representations, and ensemble strategies from blind challenge participants in our open-source model harness.
• Blind Challenge Platform: Released the full open-source codebase for our blind challenge platform template to support open science competitions across the broader research community.

Sharing Our Science

OpenADMET prioritizes the rapid dissemination of data, code, and scientific findings.

• Publications: Released 3 preprints covering our structural insights, dataset curation, and model benchmarking.
  • Structural Plasticity and Ligand Promiscuity of CYP3A4 Revealed by Cryo-EM
  • Assessing the state of the art in ADMET prediction: Lessons from the OpenADMET-ExpansionRx community blind challenge
  • Operational insights from the front lines of OpenADMET: lessons from our blind challenges
• Blog Posts & Webinars: Published 16 technical blog posts and hosted 10 interactive webinars featuring leading challenge participants and internal research teams.
• Open Data Deposition: Deposited comprehensive PXR challenge data into ChEMBL for global public access.
• Future Horizons: Looking forward to scaling our open-data engine into additional challenging Avoid-ome target classes.

Collaborative Activities

It’s exciting to see other groups using the open-source blind challenge infrastructure we developed and recently released. 

• OpenBind Blind Challenge: Launching in October, focusing on structure-only predictions for ligand binding.
• Open Free Energy Blind Challenge: Scheduled for 2027 to evaluate prospective free energy perturbation methods.

Looking Ahead

We are thrilled with the remarkable progress achieved this year and deeply grateful for the incredible response and engagement from the global research community. We have two highly unique datasets for our remaining 2026 challenges and are already mapping out the blind challenges for 2027.  

• CYP Challenge Conclusion: The CYP Blind Challenge wraps up on November 3rd.
• Phare Bio Blind Challenge: Launching in December, featuring a 10,000-compound dataset across critical antibacterial ADMET endpoints provided in partnership with Phare Bio.

Thank you for your continued support.

We would like to thank our funders for their support of OpenADMET, in particular ARPAH, Radial (part of the Astera Institute (https://ror.org/00ydx1s47)), the OpenAI Foundation, Schrödinger Inc., and the Gates Foundation. We would also like to thank our partners Phare Bio, Enamine, HuggingFace, OpenEye, CDD Vault, Discovery Life Sciences, and the beamline staff at NSLS-II for their support.

This work is supported by the Advanced Research Projects Agency for Health (ARPA-H) under AVOID-OME, and Award Number 1AY1AX000035. The contents are those of the authors. They may not reflect the policies of the Department of Health and Human Services or the U.S. government. The content is solely the responsibility of the authors and does not necessarily represent the official views of the Advanced Research Projects Agency for Health.