liveBook · Manning

Pharmacophore modeling

As explained in Build AI Drug Discovery Pipelines

Ligand-based pharmacophore modeling derives common interaction features and geometry by comparing conformations of compounds with related activity.

Pharmacophore modeling represents the interaction features and geometry associated with biological activity. A compound needs suitable functional groups, arranged at suitable distances, angles, and orientations, to interact with a biological target. Features can include hydrogen-bond donors and acceptors, hydrophobic areas, and aromatic interactions. section 7.3.2

In ligand-based modeling, the starting point is a set of ligands with related activity. For each ligand, generate reasonable conformers—possible three-dimensional arrangements—and annotate their interaction features. Align conformers from different compounds to look for shared features and geometry. The resulting pattern can help score compounds in a screening database. section 7.3.2

Figure 7.12 A) Morphine pharmacophore. B) Pharmacophore-based explanation for discrepancy in biological effects of adrenaline stereoisomers. C) Example pharmacophore features separated by a certain distance, which we can satisfy by selecting the shown functional groups and linking them to preserve desired spacing.

The examples show why feature placement matters: the orientation of an adrenaline stereoisomer affects whether it can make a hydrogen-bond contact. section 7.3.2

Structure-based modeling instead considers features in relation to a target. In fragment-based drug discovery, experimental screening identifies small fragments that bind a target; their features can inform the design and screening of larger compounds. section 7.3 section 7.3.2

One structure-based screening workflow uses density estimation to model feature placement:

  • Annotate experimentally identified fragment hits with three-dimensional pharmacophore features, such as donors, acceptors, and aromatic groups. section 7.3.3
  • Calculate distances between feature pairs across fragments and fit kernel density estimates to the observed pair distributions. section 7.3.3
  • Annotate candidate molecules and score their feature-pair distances against the fitted distributions. This is a screening aid rather than a guarantee of biological activity. section 7.3.3
Listing 7.11 Load Fragment Ensemble & Featurize with 3D Pharmacophores
import os
from tqdm import tqdm
from rdkit.Chem import MolToSmiles, PandasTools
from rdkit.Chem import ChemicalFeatures    #A
from rdkit.RDPaths import RDDataDir
 
def compute_pcore_data(mol, mol_id=0):    #B
    pcores_of_interest = \    #C
        ['Donor', 'Acceptor', 'Aromatic']    #C
 
    fdefFile = os.path.join(RDDataDir, 'BaseFeatures.fdef')
    featFactory = ChemicalFeatures.BuildFeatureFactory(fdefFile)
 
    atom_coordinates = \    #D
        mol.GetConformer().GetPositions()    #D
    pharmacophore_features = featFactory.GetFeaturesForMol(mol)
    rows = []
    for pcore in pharmacophore_features:
        pharmacophore_name = pcore.GetFamily()
        if pharmacophore_name in pcores_of_interest:
            atom_ids = pcore.GetAtomIds()
            xyz = np.mean(
                [atom_coordinates[id] for id in atom_ids], axis=0
            )
            rows.append({
                'pcore': pharmacophore_name,
                'smiles': MolToSmiles(mol),
                'mol_id': mol_id,
                'coord_x': xyz[0],
                'coord_y': xyz[1],
                'coord_z': xyz[2],
            })
    return pd.DataFrame(rows)
 
sdfFile = 'CH07/data/raw/mpro_frags.sdf'    #E
df_fragments = PandasTools.LoadSDF(
    sdfFile, idName='name', smilesName='SMILES', molColName='mol')
mols = df_fragments['mol'].values
dfs = []
for mol_id, mol in tqdm(enumerate(mols), total=len(mols)):
    dfs.append(compute_pcore_data(mol, mol_id))
pcore_df = pd.concat(dfs, ignore_index=True)
print(pcore_df.head(2))
Output:
      pcore                       smiles  mol_id  coord_x  coord_y  coord_z
0     Donor  COC(=O)c1ccc(S(N)(=O)=O)cc1       0   8.8612   5.1249   22.898
1  Acceptor  COC(=O)c1ccc(S(N)(=O)=O)cc1       0  13.0860  -1.0960   24.530

This RDKit listing extracts donor, acceptor, and aromatic features and their coordinates from fragment conformers. section 7.3.3

The density model needs enough fragment data to learn a useful distribution. Separately, when a target’s binding-site structure is available, pharmacophore hypotheses can help select an initial set of compounds for docking. section 7.3.3 section 9.2.4

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Explains ligand conformer alignment and develops a structure-based feature-distance scoring workflow using density estimation.
Examples
Illustrates feature geometry with morphine and adrenaline stereoisomers, and models feature-pair distributions from a fragment ensemble.
What you can do after
Apply it using RDKit and scikit-learn to extract fragment features, fit pair-distance densities, and score candidate molecules.
How it connects
Connects pharmacophores with fragment-based drug discovery, virtual screening, density estimation, and molecular docking.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build AI Drug Discovery Pipelines. The section links point to the parts of the book it draws on.