Towards Conditional Generation of Minimal Action Potential Pathways for Molecular Dynamics
In this paper, we utilized generative models, and reformulate it for problems in molecular dynamics (MD) simulation, by introducing an MD potential energy component to our generative model. By incorporating potential energy as calculated from TorchMD into a conditional generative framework, we attempt to construct a low-potential energy route of transformation between the helix~$\rightarrow$~coil structures of a protein. We show how to add an additional loss function to conditional generative models, motivated by potential energy of molecular configurations, and also present an optimization technique for such an augmented loss function. Our results show the benefit of this additional loss term on synthesizing realistic molecular trajectories.