Neural network reparametrization for accelerated optimization in molecular simulations

We propose a novel approach to molecular simulations using neural networkreparametrization, which offers a flexible alternative to traditional coarse-grainingmethods. Unlike conventional techniques that strictly reduce degrees of freedom,the complexity of the system can be adjusted in our model, sometimes increasingit to simplify the optimization process. Our approach also maintains continuousaccess to fine-grained modes and eliminates the need for force-matching, enhanc-ing both the efficiency and accuracy of energy minimization. Importantly, ourframework allows for the use of potentially arbitrary neural networks (e.g., GraphNeural Networks (GNN)) to perform the reparametrization, incorporating CGmodes as needed. In fact, our experiments using very weak molecular forces(Lennard-Jones potential) the GNN-based model is the sole model to find thecorrect configuration. Similarly, in protein-folding scenarios, our GNN-based CGmethod consistently outperforms traditional optimization methods. It not onlyrecovers the target structures more accurately but also achieves faster convergenceto the deepest energy states. This work demonstrates significant advancements inmolecular simulations by optimizing energy minimization and convergence speeds,offering a new, efficient framework for simulating complex molecular systems.