The instances of ('H', 0) shown in SPICE paper seem unusually low

[AndChenCM]

Hi,

I notice that there are about 1 million molecules in SPICE dataset, but only 1594 instances of ('H', 0) as shown in Table 2 in your published paper. The number '1594' seems unusually low to me. I wonder during DFT calculations, did you use explicit hydrogens for some molecules and implicit hydrogens for the rest? Or '1594' just means the explicit 'H's that are shown in smiles?

[peastman]

You're absolutely right. It's only counting hydrogens that are explicitly mentioned in the SMILES string. Here's the code I used to generate that table.

from rdkit import Chem
import h5py
from collections import defaultdict

types = defaultdict(int)
infile = h5py.File('SPICE.hdf5')
for group in infile:
    try:
        smiles = infile[group]['smiles'][0]
        mol = Chem.MolFromSmiles(smiles)
        num = infile[group]['conformations'].shape[0]
        for atom in mol.GetAtoms():
            key = (atom.GetAtomicNum(), atom.GetSymbol(), atom.GetFormalCharge())
            types[key] += num
    except:
        print(group)
for key in sorted(types):
    print(f'{key[1]}\t{key[2]}\t{types[key]}')

I just added the line

        mol = Chem.rdmolops.AddHs(mol)

to make it build implicit hydrogens, and it now gives a much more reasonable number: 15,207,949.

Sorry about that, and thanks for catching it!

[AndChenCM]

Thanks for confirming that! This makes much more sense now.