GRAPPA

We have developed GRAPPA (Graph Attentional Protein Parametrization), a machine-learned Molecular Dynamics Force Field, which is capable of predicting force-field parameters for a wide range of molecules. By combining graph-based neural networks with a transformer architecture, we achieve impressive accuracy and performance—seamlessly integrated into established MD engines such as GROMACS and OpenMM.

GRAPPA is also integrated into our reactive Molecular Dynamics (MD) pipeline KIMMDY for GROMACS, which utilizes a hybrid Kinetic Monte Carlo / MD simulation approach.

Grappa predicts MM parameters in two steps. First, atom embeddings are predicted from the molecular graph with a graph neural network. Then, transformers with symmetric positional encoding followed by permutation invariant pooling map the embeddings to MM parameters with desired permutation symmetries. Once the MM parameters are predicted, the potential energy surface can be evaluated with MM-efficiency for different spatial conformations, e.g. in GROMACS or OpenMM.

Resources

If you’d like to learn more or try out GRAPPA, you can find the software on our GitHub.

Read the detailed description of the method here.