Learning millisecond protein dynamics from what is missing in NMR spectra

The biological functions of many proteins rely on interconversions between multiple conformations occurring at microsecond to millisecond (µs-ms) timescales. A lack of standardized, large-scale experimental data has hindered obtaining a more predictive understanding of these motions. Here, after curating over 100 nuclear magnetic resonance (NMR) relaxation datasets, we found that an observable for µs-ms dynamics may have been overlooked. Millisecond dynamics can cause NMR signals to broaden beyond detection, leaving some residues not assigned in the chemical shift datasets of around 10,000 proteins deposited in the Biological Magnetic Resonance Data Bank (BMRB)¹. We made the assumption that residues with missing assignments are primarily caused by exchange-broadening due to µs-ms motions and trained various deep-learning models to predict missing assignments. These models also predict exchange measured through NMR relaxation experiments, indicative of µs-ms dynamics. The best of these models, which we named Dyna-1, leverages an intermediate layer of the multimodal language model ESM-3². Notably, dynamics directly linked to biological function, including enzyme catalysis and ligand binding, are particularly well predicted by Dyna-1, which parallels our findings that residues experiencing µs-ms exchange are more conserved. We anticipate the datasets and models presented here will be transformative in unlocking the common language of dynamics and function.