Proteins are not static objects. They move, and that motion is how they work. Enzymes flex to catalyze reactions, receptors shift to pass signals, and drugs succeed or fail depending on which states a target can reach. Yet nearly all of structural biology, from the Protein Data Bank to AlphaFold, describes proteins as single, frozen snapshots.
The era of static structure prediction has arrived, showing what's possible when quality data meets machine learning at scale. But static structures still cannot tell us what makes a designed enzyme functional, what drives drug efficacy and resistance, or how to engineer motion itself. Answering those questions requires conformational ensembles: the full set of shapes a protein adopts and the motions between them. Much of the technology to measure, model, and share ensembles still doesn't exist.
Prism is building this technology. We work across the full stack, from collecting dynamic data to modeling it with AI, encoding it in open infrastructure, and connecting it to biological function. Our goal is to make protein dynamics a well-posed AI problem and to open a dynamic era of structural biology.
Our four focus areas operate as one connected pipeline. Each advances the others.
Data collection. Making high-quality dynamic measurements routine rather than bespoke. Our initial focus is on diffuse scattering and cryo-EM.
Modeling. Using machine learning to recover the ensembles latent in experimental data.
Infrastructure. Living databases, encodings, and metrics that make ensembles usable by people and machines.
Biological impact. Tying motion to quantitative function, binding, catalysis, allostery, variant effects, at a scale where the mapping can be learned.
Nature uses motion to tune function. However, scientists mostly use shape alone, so nearly every drug and designed protein relies on shape complementarity and little else. Making motion a routine predictive variable changes that. It means predicting how a disease mutation shifts binding, stability, or activity rather than only how it alters a structure. It means designing drugs against specific functional states, including transient pockets that never appear in a static model, with real selectivity between closely related targets. It means engineering motion into enzymes for synthetic biology. It also means being able to ask questions about molecular function that the field has not yet learned to pose.
This work has to be coordinated: shared standards, open data, common file formats, benchmarks, and infrastructure that lets one group’s measurements become another group’s training data. Prism is a fully collaborative project bringing together scientists from Radial and leading academic institutions. To accelerate the science we are doing, all of our work is fully open. We publish data, methods, code, and reasoning continuously and outside traditional journals, and we aim for everything we release to be findable, accessible, interoperable, and reusable.
Learn more about Prism