Deployment

Installation

pip install rai-chem

Usage

This package can be utilised to:

• extract useful atom/subgroup features from small molecules and proteins (e.g. hybridization, valence, hydrophobicity, donor/acceptor properties, electrostatic properties)
• detect intermolecular non-covalent interactions (e.g. hydrophobic, electrostatic, hydrogen, halogen bonds, pi stacking, amide-pi, cation-pi)
• detect unfavorable atom-atom contacts (e.g. steric clashes, donors/acceptors/cations/anions in proximity)

We recommend to add the hydrogens to protein PDB files as it is vital for the hydrogen bonds detection. If your input files don't include H atoms, you may use reduce package or any other tool of your choice.

To extract and store features of a small molecule atoms (lig.atoms), aromatic rings (lig.aromatics), and amide groups (lig.amides):

from rdkit import Chem
from rai_chem.ligand import Ligand

mol = next(Chem.SDMolSupplier('my_ligand.sdf', removeHs=False))
mol = Chem.AddHs(mol, addCoords=True)  # if SDF file doesn't contain Hs

lig = Ligand('my_ligand', mol)

To extract and store features of a protein atoms (prot.atoms), aromatic rings (prot.aromatics), and amide groups (prot.amides):

from rai_chem.protein import PDBParser, Protein

with open('my_protein.pdb', "r") as f:
    pdb_lines = f.readlines()
prot_table = PDBParser('my_protein', pdb_lines, remove_hs=False)
prot = Protein('my_protein', prot_table.atoms)

To detect protein-ligand non-covalent interactions (pli.interactions, pli.to_table()) and unfavorable contacts (pli.unfavorable):

from rai_chem.interaction import ProtLigInteractor

pli = ProtLigInteractor(prot, lig)
df_interactions = pli.to_table()

The scoring of small molecule poses within protein binding site can be performed with:

python -m rai_chem.score --pdb my_protein.pdb --sdf my_ligands.sdf --save_path ligands_pose_score.sdf

The scores in the output SDF file include:

• FavorableRate: total number of favorable interactions normalised by the number of ligand heavy atoms (the contribution of the hydrophobic interactions was accounted for with a weight of 1/6 due to their high relative abundance);
• FavorableRateUniq: fraction of the heavy ligand atoms participating in at least one non-bond interaction;
• UnfavorableRate: total number of unfavorable interactions normalised by the number of ligand heavy atoms;
• UnfavorableRateUniq: fraction of the ligand heavy atoms participating in at least one unfavorable interaction;
• VDWClash: sum of all clash distances, which are below the steric clashes threshold;
• FitScore: FavorableRate + FavorableRateUniq - 2UnfavorableRate - 2UnfavorableRateUniq - 10*VDWClash

Cite

This package was created and updated as part of the following works within RECEPTOR.AI:

• Augmenting a training dataset of the generative diffusion model for molecular docking with artificial binding pockets (DOI: doi.org/10.1039/D3RA08147H)

• 3DProtDTA: a deep learning model for drug-target affinity prediction based on residue-level protein graphs (DOI: doi.org/10.1039/D3RA00281K)

Please cite the above-mentioned preprints/articles if you use this code in your own work.

Acknowledgements

This package is inspired by the Open Drug Discovery Toolkit and based on the RDKit