Charges dependent on the conformation

[maciejwisniewski-drugdiscovery]

Hi,
Are the Espaloma charges dependent on the conformation of the chemical compound?
Because after using this code, I get the same charges for each conformation, and I thought that the AM1 method inherently takes conformation into account, unlike, for example, Gasteiger.

Code:

import rdkit
from rdkit import Chem
from rdkit.Chem import AllChem
from rdkit.Chem import rdDistGeom, rdFMCS
from openff.toolkit import Molecule
from espaloma_charge.openff_wrapper import EspalomaChargeToolkitWrapper
import numpy as np
import time
from espaloma_charge import charge

toolkit_registry = EspalomaChargeToolkitWrapper()

def get_conformations(molecule):
    all_partial_charges = []
    molecule.RemoveAllConformers()
    molecule = Chem.AddHs(molecule)
    params = rdDistGeom.ETKDGv3()
    params.randomSeed = 0xd06f00d
    params.numThreads = 4
    params.maxAttempts = 100
    # print how many atoms there is
    cids = rdDistGeom.EmbedMultipleConfs(molecule, 32, params)
    if len(cids) == 0:
        print("Failed to generate conformations for molecule. Skipping.")
        return None, None
    molecule = Chem.RemoveHs(molecule)
    # Calculating Partial Charges with AM1-BCC
    conformers = Molecule.from_rdkit(molecule,
                                     allow_undefined_stereo=True,
                                     hydrogens_are_explicit=False).conformers
    molecule.RemoveAllConformers()
    for conformer in conformers:
        start = time.time()
        template = Molecule.from_rdkit(molecule,
                                       allow_undefined_stereo=True,
                                       hydrogens_are_explicit=False)
        template.add_conformer(conformer)
        
        template.assign_partial_charges(
            partial_charge_method = 'espaloma-am1bcc',
            toolkit_registry=toolkit_registry,
            use_conformers = template.conformers
        )
        end = time.time()
        print('DURATION:\t',(end - start)/60)
        all_partial_charges.append(template.partial_charges)
        
    return all_partial_charges

rdkit_mol = Chem.MolFromSmiles('COC(=O)C(C1CCCCN1)C2=CC=CC=C2')
charges = get_conformations(rdkit_mol)

[ljmartin]

I wondered this too - looks like the graph definition does not include conformer information:

espaloma-charge/espaloma_charge/utils.py

Line 79 in 6c9fec8

def from_rdkit_mol(mol, use_fp=True):

[jgreener64]

Espaloma charges do not depend on conformation, just topology, though Espaloma was trained on conformation-dependent charges (I am not involved in the project).

[pavankum]

The paper mentions training to AM1-BCC ELF10 charges, which is a method to diminish conformer dependence, https://pubs.acs.org/doi/full/10.1021/acs.jpca.4c01287.