BLISS Speaker Series Episode #18

BLISS is pleased to welcome Dr Yu Xie, Senior Research Scientist at Microsoft AI4Science, who will give a 45-minute talk titled ‘Scalable Emulation of Protein Equilibrium Ensembles with Generative Deep Learning’. After the talk, attendees will have the opportunity to mingle with other AI enthusiasts, share ideas, ask questions and enjoy complimentary drinks. Please note that doors will close promptly at 7:15pm, so early arrival is highly recommended.

An RSVP via Meetup is essential to guarantee entry. Although Meetup has been actively promoting its Plus programme recently, there is no obligation to purchase it - both the platform and all BLISS events remain completely free.

Abstract:
Following the sequence and structural revolutions, predicting the dynamic mechanisms of proteins that control biological functions remains an outstanding scientific challenge. While experimental techniques and molecular dynamics (MD) simulations can in principle determine conformational states, binding configurations and their probabilities, they suffer from low throughput.
In this talk, Dr Xie will present BioEmu, a biomolecular emulator - a generative deep learning system capable of generating thousands of statistically independent samples from the protein structure ensemble per hour on a single GPU. Using novel training techniques, extensive protein structure data, MD simulations with a duration of over 200 milliseconds and experimental protein stability data, BioEmu generates equilibrium ensembles that excel in a range of sophisticated and practically relevant metrics.

Qualitatively, BioEmu captures a broad spectrum of functionally significant conformational changes, including cryptic pocket formation, unfolding of specific regions and large-scale domain rearrangements. Quantitatively, it achieves the detection of protein conformations with relative free energy errors of approximately 1 kcal/mol, validated using millisecond MD simulations and experimental data. By mimicking structural ensembles and thermodynamic properties, BioEmu provides mechanistic insights into protein behaviour - such as understanding fold destabilisation in mutants - and supports the generation of experimentally testable hypotheses.