Accelerating Protein Dynamics Research: MDGen Now Available on Vecura

What is MDGen?

MDGen is a trajectory-generative deep learning model designed to predict future protein conformations from an initial structure. By leveraging a flow-matching generative approach, it learns patterns from structural ensemble trajectories to simulate protein dynamics without the need for computationally expensive, physics-based explicit-solvent molecular dynamics. It allows users to produce atomically detailed conformational ensembles in seconds rather than days.

It helps users explore the dynamic behavior of protein monomers, such as loop fluctuations and hinge bending, providing a rapid way to assess conformational mobility. It is especially useful for structural biologists and researchers who need a preliminary look at protein motion or a diverse starting set of frames to seed downstream, more intensive physics-based MD simulations.

What can users do with MDGen on Vecura?

With MDGen on Vecura, users can:

• Generate atomically detailed, MD-style trajectories for single-chain protein monomers.
• Create multi-frame conformational ensembles starting from a single PDB structure.
• Chain rollouts to extend the total duration of the simulated trajectory.
• Export results in PDB format or as binary XTC trajectory files for use in industry-standard analysis tools like MDAnalysis, VMD, and PyMOL.

What the output means

The output provides a series of predicted protein conformations in the form of a multi-MODEL PDB file or a binary XTC trajectory file, paired with a topology file. The generated ensemble represents plausible conformational states that statistically mirror those produced by explicit-solvent MD.

This output should be used to support scientific decision-making, such as identifying areas of interest for further study. It does not replace experimental validation or quantitative thermodynamic analysis performed through traditional physics-based methods.

Why this matters

Molecular dynamics simulations are essential for understanding protein function and stability, but they are traditionally extremely computationally intensive, often requiring large GPU clusters and significant time. MDGen democratizes this access by providing a deep-learning-based alternative that captures structural dynamics with remarkable speed.

By enabling researchers to perform high-throughput conformational sampling in seconds, MDGen lowers the barrier to investigating protein dynamics. While it does not replace the precision of force-field-based thermodynamics, it serves as a powerful screening tool to prioritize targets and guide more accurate, compute-heavy simulations.

• Developed by: Researchers at MIT (bjing2016)
• Source: GitHub Repository (https://github.com/bjing2016/mdgen)
• Reference: arXiv:2409.17808 (https://arxiv.org/abs/2409.17808)