Potential contrasting is an efficient method for learning a potential energy function that can reproduce an ensemble of molecular conformations. It can be easily applied to can learn coarse-grained force fields based on all-atom simulations. It generalizes the noise contrastive estimation method to use complex unnormalized noise distributions constructed using molecular dynamics techniques such as umbrella sampling.
Need help? Checkout the documentation.