Machine Learning of Collective Variables for Accelerating Molecular Dynamics Simulations

Molecular dynamics simulations are an invaluable tool in various fields of science, with hundreds of applications in material science, drug discovery, protein folding, and more.

However, in many interesting cases, a direct simulation of the physical dynamics is computationally infeasible, or very expensive.

We're developing novel algorithms, combining deep learning and enhanced sampling techniques, to accelerate simulations in such cases, especially in cases where the interesting bit of the physics is dominated by entropy.