Impact of 'radius' on GetMorganFingerprint()

Cheminformatics Tutorials - Herong's Tutorial Examples

∟Impact of 'radius' on GetMorganFingerprint()

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

The 'radius' option in the rdkit.Chem.rdMolDescriptors.GetMorganFingerprint() function call allows you to control the maximum radius of local substructures to be encoded into the fingerprint.

1. For example, molecule "CCC" will produce 3 identifiers with radius=0 in the fingerprint. 1 identifier for each atom node. But atom 0 and atom 2 will have the same identifier, since they have the same number of connected bonds.

from rdkit.Chem import AllChem
from rdkit.DataStructs.cDataStructs import UIntSparseIntVect
radius = 0
bitInfo = {}
mol = AllChem.MolFromSmiles('CCC')
fp = AllChem.GetMorganFingerprint(mol, radius, bitInfo=bitInfo)
display(UIntSparseIntVect.GetLength(fp))
display(UIntSparseIntVect.GetNonzeroElements(fp))
display(UIntSparseIntVect.GetTotalVal(fp))
print(bitInfo)

# output: 
4294967295
{2245384272: 1, 2246728737: 2}
3
{2245384272: ((1, 0),), 2246728737: ((0, 0), (2, 0))}

2. But the same molecule "CCC" will produce 6 identifiers with radius=1. 3 of them identifies 3 atoms with atom invariants encoded only corresponding to radius=0 output. The other 3 identifies are updated identifies from radius=0 with information from neighboring nodes with radius=1 (1 bond away from the atom).

radius = 1
bitInfo = {}
mol = AllChem.MolFromSmiles('CCC')
fp = AllChem.GetMorganFingerprint(mol, radius, bitInfo=bitInfo)
display(UIntSparseIntVect.GetLength(fp))
display(UIntSparseIntVect.GetNonzeroElements(fp))
display(UIntSparseIntVect.GetTotalVal(fp))
print(bitInfo)

# output:
4294967295
{2068133184: 1, 2245384272: 1, 2246728737: 2, 3542456614: 2}
6
{2068133184: ((1, 1),), 
 2245384272: ((1, 0),), 
 2246728737: ((0, 0), (2, 0)), 
 3542456614: ((0, 1), (2, 1))}

We can read the output as:

"2068133184" identifies atom 1 with radius 1, or a "C" atom with 2 neighboring "C" atoms at 1 bond distance.
"2245384272" identifies atom 1 with radius 0, or a "C" atom with 2 non-Hydrogen bonds.
"2246728737" identifies atom 0 and atom 2 with radius 0, or a "C" atom with 1 non-Hydrogen bonds.
"3542456614" identifies atom 0 and atom 2 with radius 1, or a "C" atom with 1 neighboring "C" atom at 1 bond distance.

As you can see, the Morgan generator keeps not only the updated identifiers from the last iteration of the Morgan algorithm, but it also keeps identifiers from intermediate iterations.

3. To try radius=2, we need to use a larger molecule "CCCC" This time, total of 9 identifiers are produced. 3 identifiers of each of 3 iterations: radius=0, radius=1, and radius=2.

radius = 2
bitInfo = {}
mol = AllChem.MolFromSmiles('CCCC')
fp = AllChem.GetMorganFingerprint(mol, radius, bitInfo=bitInfo)
display(UIntSparseIntVect.GetLength(fp))
display(UIntSparseIntVect.GetNonzeroElements(fp))
display(UIntSparseIntVect.GetTotalVal(fp))
print(bitInfo)

# output: 
4294967295
{1173125914: 2, 1244535424: 1, 2245384272: 2, 2246728737: 2, 3542456614: 2}
9
{1173125914: ((1, 1), (2, 1)), 
 1244535424: ((1, 2),), 
 2245384272: ((1, 0), (2, 0)), 
 2246728737: ((0, 0), (3, 0)), 
 3542456614: ((0, 1), (3, 1))}

Conclusion: higher value of radius produces more identifiers in the final fingerprint. The total number of identifiers is exactly the number of non-Hydrogen atom nodes times radius+1.