Dataset of bond dissociation energies

This directory consists of the data files needed to train BonDNet. Note, further processing might be needed to extract the necessary information, since this is not formatted the same as what BonDNet expects. See the below link for what BonDNet expects: https://github.com/mjwen/bondnet/tree/master/bondnet/scripts/examples/train

Files

1. molecules.sdf All molecules in the dataset are stored in this file. The file is in SDF format, which can be read by RdKit, Open Babel or other cheminformatics tools.

2. molecule_attributes.yaml This consists of additional attributes for each molecule in the dataset. Note, they are not used in the BonDNet paper.

3. reactions.yaml This file consists of all the bond dissociation reactions, a reaction may look like this:

- atom_mapping:
    - 0: 0
      1: 2
      2: 1
    - 0: 3
  bond_mapping:
    - 0: 2
      1: 0
      2: 3
    - {}
  id: 5f07dfcc578f819b3d12721e_broken_bond-0-3_species-C-H
  index: 1
  products:
    - 8251
    - 97
  reactants:
    - 7943
  value: 4.331072762383323

Here are the key fields:

• reactants: ID of the reactant molecules (in the molecules.sdf file)
• products: ID of the product molecules (in the molecules.sdf file)
• value: The bond dissociation energy of the reaction in eV
• atom_mapping: The mapping of atoms in the reactants and products. The first index is the reactant index, the second index is the product index.
• bond_mapping: The mapping of bonds in the reactants and products. The first index is the reactant index, the second index is the product index.

Reference

1. Wen, Mingjian, Samuel M. Blau, Evan Walter Clark Spotte-Smith, Shyam Dwaraknath, and Kristin A. Persson. "BonDNet: a graph neural network for the prediction of bond dissociation energies for charged molecules." Chemical science 12, no. 5 (2021): 1858-1868.