Molecular dynamics (MD) simulations are a cornerstone in the study of the thermodynamics and kinetics of materials, biomolecules, and chemical reactions. Yet, their scope is inherently limited in terms of length and time scales. Despite the prowess of advanced computational platforms, the simulation of intricate biomolecular systems and materials often doesn't exceed milliseconds. This limitation arises primarily because molecules in MD simulations tend to get ensnared in stable or metastable states.

Over time, the quest to elongate the time frame of these simulations has given rise to a plethora of enhanced sampling techniques. The crafting of these techniques has been informed by a fusion of physical laws, chemical insight, and data science tools.

The advent of contemporary data science has ushered in an array of methods to glean insights from MD simulation trajectories. These insights can then be looped back into the simulation to aid in the sampling of diverse molecular conformations. Our research interest lies in harnessing machine learning to devise a compact model for intricate biomolecules. This includes the formulation of machine learning-driven collective variables and the creation of machine-learning-based coarse-grained models. Further, we are keen on utilizing machine learning to predict protein configurations using all-atom MD simulations.