Machine learning coarse-grained potentials of protein thermodynamics

Molecular Dynamic simulations are becoming a standard tool to characterize protein dynamics. Despite the significant progress and impressive results obtained utilizing atomic-level force fields, simulations longer than milliseconds are still unfeasible for large protein systems. We present recent results on the developments of protein models at reduced resolution that can still well reproduce a protein configurational landscape but at a small fraction of the cost. We discuss the method development and the results, and we present a quantitative comparison of the thermodynamic and dynamic properties of different macromolecular systems at different resolutions.