Bond featurizers

from chemprop.featurizers.bond import MultiHotBondFeaturizer

This is an example bond to featurize.

from rdkit import Chem

bond_to_featurize = Chem.MolFromSmiles("CC").GetBondBetweenAtoms(0, 1)

Bond features

The following bond features are generated by rdkit and cast to one-hot vectors (except for the initial null bit which is True/False depending on if the bond is None). These feature vectors are joined together to a single multi-hot feature vector. Only the stereochemistry vector is padded for unknown values.

• null?

• bond type

• conjugated?

• in ring?

• stereochemistry

featurizer = MultiHotBondFeaturizer()
featurizer(bond_to_featurize)

array([0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0])

Custom

The bond types and stereochemistry can be customized. The defaults are:

• bond_type

  • Single, Double, Triple, Aromatic

• stereos

  • 0, 1, 2, 3, 4, 5 - See rdkit.Chem.rdchem.BondStereo for more details

from rdkit.Chem.rdchem import BondType

featurizer = MultiHotBondFeaturizer(bond_types=[BondType.SINGLE], stereos=[0, 1, 2])
featurizer(bond_to_featurize)

array([0, 1, 0, 0, 1, 0, 0, 0])

Generic

Any class that has a length and returns a numpy array when given an rdkit.Chem.rdchem.Bond can be used as a bond featurizer.

from rdkit.Chem.rdchem import Bond
import numpy as np


class MyBondFeaturizer:
    def __len__(self):
        return 1

    def __call__(self, a: Bond):
        return np.array([a.GetIsConjugated()], dtype=float)


featurizer = MyBondFeaturizer()
featurizer(bond_to_featurize)

array([0.])