Most medicines work by binding to disease-related proteins, and predicting which molecules will bind, and how strongly, is central to early-stage drug design. While AI has transformed protein structure prediction, its impact on binding prediction has been held back by a shortage of consistent experimental data. The OpenBind consortium, which includes University of Oxford researchers, aims to fix that by building the world's largest open dataset of protein-ligand interactions.
The consortium's first release provides detailed X-ray structures of 699 compounds binding to the EV-A71 virus protein, with binding strength measurements for 601 of them, one of the largest public datasets ever assembled for a single protein target. The data were generated through high-throughput pipelines at Diamond Light Source in Oxfordshire, combining automated chemistry, robust binding measurements and crystallography, processed into formats designed for machine learning, and accompanied by a predictive AI model.
Even the most advanced systems in structural biology, such as AlphaFold and Boltz, are limited by their training data and struggle with targets unlike anything they have seen. By generating large volumes of new, standardised experimental measurements, OpenBind intends to push models beyond pattern-matching towards reliable predictions about novel compounds, streamlining a discovery process in which narrowing down viable candidates remains slow and expensive.