Incorrect detection of interactions

[DrrDom]

I observed multiple incorrect contact detection, which have a similar pattern. The problem was observed starting from version 2.0.X (do not remember the exact number) and till the latest version 2.2.0.

from rdkit import Chem
import prolif as plf

protein_pdb_file = 'XXX'
ligand_sdf_file = 'YYY'

protein_mol = Chem.MolFromPDBFile(
    protein_pdb_file,
    removeHs=False,
    sanitize=True,
)

ligand_mol = Chem.SDMolSupplier(ligand_sdf_file, removeHs=False, sanitize=True)[0]

ligand = plf.Molecule(ligand_mol)
protein = plf.Molecule(protein_mol)
fingerprint = plf.Fingerprint(interactions="all", count=True)
ifp = fingerprint.generate(
    ligand,
    protein,
    residues=residue_filter,
    metadata=True,
)

If look at Leu106 residue, there is HBDonor contact between OH group of a ligand and NH group of a protein. That is definitely incorrect.

ChatGPT suggests that this may happen due to the split of protein on individual residues and afterwards applying SMARTS to detect interaction points. Therefore, NH becomes a valid H-bond acceptor partner.

I attached the test files

test_files.zip

[cbouy]

That's embarrassing, I actually noticed this bug more than a year ago and I remember putting it on a TODO list to just add to the SMARTS pattern the "fragmented" backbone nitrogen (which doesn't match the usual !$([NX3]-*=[O,N,P,S]) pattern), but then I wiped the laptop that had the TODO list and completely forgot about it...

Anyway, thanks for finding this, adding &!$([ND2v3^2+0](-[H])-[CD4v4H1^3]-[CD2^2+0]=O) to the current acceptor SMARTS seems to do the trick and should hopefully be specific enough to not exclude other unrelated nitrogen atoms.

I'll leave the issue open to remind me to work on the proper fix (doing the substructure matching on the unfragmented version of the molecule), but this will require a substantial refactor.

[DrrDom]

Thank you for the quick fix! Now this issue was gone if ProLIF is installed from the master branch.