A major scientific advance in protein modeling, developed by Microsoft Research AI for Science, has been published in the July 10 issue of Science. The study introduces BioEmu, a generative deep learning system that emulates the equilibrium behavior of proteins with unprecedented speed and accuracy. As the biological function of proteins depends on dynamical changes in their structure, the ability to predict these structural changes quickly and accurately opens the door to more rational design decisions in drug discovery, helping to reduce the failure rate of drugs in clinical trials.

BioEmu can generate thousands of statistically independent protein structures per hour on a single graphics processing unit (GPU). “This reduces the cost and the required time to analyze functional structure changes in proteins,” says Professor Frank Noé who led the project. BioEmu integrates over 200 milliseconds of molecular dynamics simulations with experimental data to predict structural ensembles and thermodynamic properties with near-experimental accuracy.

The system captures complex biological phenomena such as the formation of hidden binding pockets, domain motions, and local unfolding — all critical to understanding protein function and drug design. BioEmu also predicts protein stability changes with an accuracy that can compete with laboratory experiments. “Thereby, BioEmu provides a scalable method to model protein function at the genomic scale,” adds Professor Cecilia Clementi.

The BioEmu code and model are freely available under the permissive MIT license. Alongside the publication, Microsoft Research has also released the molecular dynamics simulation dataset that was generated to train BioEmu. This dataset – comprising over 100 milliseconds of simulations across thousands of protein systems – represents the largest sequence-diverse protein simulation set publicly available to date.

While the research was conducted entirely at Microsoft, Freie Universität Berlin is proud to acknowledge the contributions of affiliated researchers. The research was led by Frank Noé, Partner Research Manager at Microsoft Research AI for Science in Berlin, who also holds an honorary professorship at Freie Universität Berlin. Cecilia Clementi, Einstein Professor for Theoretical and Computation Biophysics at Freie Universität Berlin, made key contributions to the work as a visiting researcher at Microsoft Research.

The study, “Scalable Emulation of Protein Equilibrium Ensembles with Generative Deep Learning,” published in Science, is available online at: https://doi.org/10.1126/science.adv9817.