Impact of 'invariants' on GetMorganFingerprint()

Cheminformatics Tutorials - Herong's Tutorial Examples

∟Impact of 'invariants' on GetMorganFingerprint()

This section provides a tutorial example on impact of the 'invariants' option on fingerprint generation with rdkit.Chem.rdMolDescriptors.GetMorganFingerprint() function.

The 'invariants' option in the rdkit.Chem.rdMolDescriptors.GetMorganFingerprint() function call allows you to override initial identifiers hashed from those predefined invariants (descriptors) on all atom nodes. Basically, you are providing the Morgan fingerprint for radius=0.

By default, RDKit uses the same invariants on each atom node as defined by Daylight Chemical Information Systems Inc..

Number of non-hydrogen immediate neighbors ( also called heavy atom connections).
Valency minus the number of connected hydrogens.
Atomic number of atom at the node.
Atomic mass of atom at the node.
Atomic charge of atom at the node.
Number of Hydrogens connected to the node.
Ring membership of the node.

1. For example, the following code shows you the differences between default initial identifiers and user-provided initial identifiers:

from rdkit.Chem import AllChem
mol = AllChem.MolFromSmiles('CC')

bitInfo = {}
fp = AllChem.GetMorganFingerprint(mol, 1, invariants=[], bitInfo=bitInfo)
print(bitInfo)

bitInfo = {}
fp = AllChem.GetMorganFingerprint(mol, 1, invariants=[1,2], bitInfo=bitInfo)
print(bitInfo)

# output: 
{2246728737: ((0, 0), (1, 0)), 3545175291: ((0, 1),)}
{1: ((0, 0),), 2: ((1, 0),), 3205495808: ((0, 1),)}

As you can see from the output, when initial identifiers are provided, Morgan generator returns them as is as radius=0 identifiers, and uses them to derive radius=1 identifiers.

2. You can actually retrieve default initial identifiers from a given molecule by calling the GetConnectivityInvariants(mol) method. Them provide them the GetMorganFingerprint() call. You will get the same fingerprint as the default invariants.

mol = AllChem.MolFromSmiles('CC')
bitInfo = {}
inv = AllChem.GetConnectivityInvariants(mol)
print(inv)
fp = AllChem.GetMorganFingerprint(mol, 1, invariants=inv, bitInfo=bitInfo)
print(bitInfo)

bitInfo = {}
fp = AllChem.GetMorganFingerprint(mol, 1, invariants=[], bitInfo=bitInfo)
print(bitInfo)

# output:
[2246728737, 2246728737]
{2246728737: ((0, 0), (1, 0)), 3545175291: ((0, 1),)}
{2246728737: ((0, 0), (1, 0)), 3545175291: ((0, 1),)}

Conclusion: The "invariants=[...]" option is not easy to use. You have to build your own initial identifier on each atom node, by collecting a set of invariant values and hashing them into an integer.

If you really want to build your own initial identifiers, you can follow this example based on the code provided by Greg Landrum at mail-archive.com/rdkit-discuss@lists.sourceforge.net/msg09400.html.

from rdkit import Chem
from rdkit.Chem import AllChem

def generateECFPAtomInvariant(mol, discrete_charges=False):
    pt = Chem.GetPeriodicTable()
    num_atoms = mol.GetNumAtoms()
    invariants = [0]*num_atoms
    ring_info = mol.GetRingInfo()
    for i,a in enumerate(mol.GetAtoms()):
        descriptors=[]
        descriptors.append(a.GetAtomicNum())
        descriptors.append(a.GetTotalDegree())
        descriptors.append(a.GetTotalNumHs())
        descriptors.append(a.GetFormalCharge())
        descriptors.append(a.GetMass() - pt.GetAtomicWeight(a.GetSymbol()))
        if(ring_info.NumAtomRings(i)):
            descriptors.append(1)
        invariants[i]=hash(tuple(descriptors))& 0xffffffff
    return invariants

mol = Chem.MolFromSmiles('C')
display(generateECFPAtomInvariant(mol))
display(AllChem.GetConnectivityInvariants(mol))

# output: 
[2286409670]
[2246733040]

The generateECFPAtomInvariant() code is not generating the same identifier as AllChem.GetConnectivityInvariants(). Maybe the "discrete_charges=True" option needed to be added.