Mapping Transporters and Blood-Brain Barrier Penetration with the OpenAI Foundation

Every small-molecule drug discovery program eventually faces the issue of how compounds distribute throughout the body. You can optimize binding affinity to picomolar potency, but if the molecule cannot reach its intended site of action, it will never become a medicine.

Nowhere is this challenge more evident than at the blood-brain barrier (BBB). In Alzheimer's disease alone, more than 200 compounds have entered clinical trials over the past two decades, with a 99.6% failure rate. A major cause of these failures is active efflux: membrane transporters such as P-glycoprotein (P-gp) recognize small molecules and actively pump them back out of the central nervous system.

Computational prediction of these interactions remains challenging because public in vitro transport and physicochemical data are sparse, siloed, and often measured with inconsistent protocols across labs.

We are excited to announce that OpenADMET has partnered with the OpenAI Foundation, through its Public Data for Health program, to generate large-scale, physically grounded in vitro datasets covering physicochemical properties and transporter biology.

What We Are Building

To build machine learning models that accurately predict transport liabilities and compound disposition, we need systematic, standardized in vitro ground truth. This initiative brings together the core strengths of the OpenADMET consortium:

• High-Throughput In Vitro Profiling: Octant will use its high-throughput analytical chemistry pipelines to profile physicochemical properties and transporter interactions across tens of thousands of diverse small molecules.
• Transporter Structural Biology: Structural biologists at UCSF will determine novel, high-resolution structures of key transporters that govern passage across biological barriers.
• Open Datasets and Blind Benchmarks: We will make these datasets publicly available and run prospective community blind challenges to benchmark predictive models against newly measured experimental data.

Expanding OpenADMET's Scope

Small molecules remain the dominant modality in medicine because they can be taken orally, are relatively easy to synthesize, and can reach difficult biological compartments. Providing open access to high-quality in vitro transporter data and structural insights will help the community design better molecules with fewer late-stage surprises.

We are deeply grateful to the OpenAI Foundation for supporting this initiative through its Public Data for Health program.

Stay tuned for upcoming dataset releases, structural biology updates, and blind challenges at openadmet.org.

OpenADMET is supported by ARPA-H (under the AVOID-OME program, Award Number 1AY1AX000035), Radial and the Astera Institute (https://ror.org/00ydx1s47), Schrödinger, the Bill & Melinda Gates Foundation, and the OpenAI Foundation. We also thank our partners Enamine, Hugging Face, OpenEye, CDD Vault, Discovery Life Sciences, and the NSLS-II beamline staff for their ongoing support.