QPredict: Partial charge prediction using AI/ML NFP model (5:48)

This workflow illustrates how to use a trained graph neural network (GNN) model to predict the partial charges on organic molecules. These particular saved models have been trained on a dataset of ~40000 high-fidelity DFT calculations. The GNN model has been constructed using the neural fingerprint (NFP) package developed at NREL.

Benefits of GNN models include:
  • one can localize predictions to subsections of a molecule

  • in general, these models outperform fingerprint-based methods

  • accuracy usually improves with added data, while fingerprint-based methods tend to saturate in performance.

Full 1080p HD resolution and full-screen available through menu on the bottom right of the video