README.md

Please note that this repository is published for transparency on how the datasets were generated. It is not meant as codebase for generating custom datasets or for running QM calculations.

Instructions for building grappa datasets and what is required as input can be found at our Google Colab Tutorial.

Grappa Data Creation

Scripts used to create the peptide datasets 'uncapped_amber99sbildn', 'tripeptides_amber99sbildn', 'hyp-dop_amber99sbildn' and 'dipeptide_rad' from grappa-1.1.

For all of these, the workflow was:

• Generate PDB files from sequences using the pepgen package (generate_pdbs.py)
• Sample states from MD at 300K using a traditional forcefield (amber99sbildn for peptides, an early version of grappa for radicals). To reach a larger conformational space, we simulate at 500K and reduce the temperature back to 300K before sampling a state (generate_states.py)
• Calculate single point energy and gradient with DFT (single_points.py)
• Save energy, gradient, sequence, pdb-string and atomic numbers as npz file that can be read by grappas data.MolData class (to_grappa_format.py)