RDKit Conversion Issue

[zachcp]

Hi @padix-key ,

As a follow up to zachcp/moltite#1, I just tried to use this new feature and the conversion gives me an empty stack. I can track this down if needed but do you think you could confirm if this is reproducible on your end?

import rdkit
import biotite

print(biotite.__version__)
print(rdkit.__version__)
# 1.1.1.dev43+g4c7204af
# 2024.09.5


from rdkit import Chem
from biotite.interface import rdkit as irdk

# Example SMILES string
smiles = "CC(=O)OC1=CC=CC=C1"  # Aspirin

# Convert SMILES to mol object
mol = Chem.MolFromSmiles(smiles)

biotite_mol = irdk.from_mol(mol)
biotite_mol

# the conversion gives me an empty stack...
# stack([ ])

[zachcp]

I can confirm the above with this minimal env. This is python 3.13.

# pixi.toml
[project]
authors = ["Zachary Charlop-Powers <>"]
channels = ["conda-forge"]
name = "test"
platforms = ["osx-arm64"]
version = "0.1.0"

[tasks]

[dependencies]
rdkit = ">=2024.9.5,<2025"
ipython = ">=8.32.0,<9"

[pypi-dependencies]
biotite = { git = "https://github.com/biotite-dev/biotite", rev = "4c7204afd07c489117243fd8ad4a67f787a4a820" }

pixi run ipython
# paste code from above.

[padix-key]

Hi, thanks for reporting. @Croydon-Brixton also encountered this problem (#741 (comment)) and already prepared a fix (padix-key#13). The reason is that the AtomArrayStack contains one model per conformer and a Mol freshly created from a SMILES string does not contain any conformer. In the fix from @Croydon-Brixton the AtomArrayStack will have a coordinate array full of NaN values in this case.

To get actual coordinates you have to predict them and (optionally) add hydrogen atoms.
For example like here (taken from the tutorial in the main branch):

    from rdkit.Chem import MolFromSmiles
    from rdkit.Chem.rdDistGeom import EmbedMolecule
    from rdkit.Chem.rdForceFieldHelpers import UFFOptimizeMolecule
    from rdkit.Chem.rdmolops import AddHs

    ERTAPENEM_SMILES = "C[C@@H]1[C@@H]2[C@H](C(=O)N2C(=C1S[C@H]3C[C@H](NC3)C(=O)NC4=CC=CC(=C4)C(=O)O)C(=O)O)[C@@H](C)O"

    mol = MolFromSmiles(ERTAPENEM_SMILES)
    # RDKit uses implicit hydrogen atoms by default, but Biotite requires explicit ones
    mol = AddHs(mol)
    # Create a 3D conformer
    conformer_id = EmbedMolecule(mol)
    UFFOptimizeMolecule(mol)
    ertapenem = rdkit_interface.from_mol(mol, conformer_id)

[zachcp]

Beautiful. Thank you.