Impact of 'fpSize' on RDKFingerprint()

Cheminformatics Tutorials - Herong's Tutorial Examples

∟Daylight Fingerprint Generator in RDKit

∟Impact of 'fpSize' on RDKFingerprint()

This section provides a tutorial example on impact of the 'fpSize' option on fingerprint generation with rdkit.Chem.rdmolops.RDKFingerprint() function.

The 'fpSize' option in the rdkit.Chem.rdmolops.RDKFingerprint() function call allows you to control the size (length) of the fingerprint. The default is fpSize=2048, which is capable to support 1024 subgraphs with nBitsPerHash=2 in a perfect scenario.

Generating shorter fingerprints can save data storage. But it also reduces the accuracy the final fingerprints.

1. For example, "CCCCCCCC" and "OOOOOOOO" are 2 completely different molecules with a zero similarity between them. But their fingerprints share 3 common bits, when fpSize=16 is used. And that gives a 0.27 similarity score.

from rdkit import Chem
from rdkit import DataStructs
mol = Chem.MolFromSmiles('CCCCCCCC')
fp1 = Chem.RDKFingerprint(mol, fpSize=16, nBitsPerHash=1)
print(fp1.ToBitString())

mol = Chem.MolFromSmiles('OOOOOOOO')
fp2 = Chem.RDKFingerprint(mol, fpSize=16, nBitsPerHash=1)
print(fp2.ToBitString())

print(DataStructs.FingerprintSimilarity(fp1, fp2))

# output 
0100110001011100
1100001001110010
 ^       ^ ^
0.2727272727272727

The main reason for two different subgraphs turning on the same bit in the fingerprint is that the fingerprint is not long enough to allow them to use different bits.

2. Now if we use "fpSize=64", "CCCCCCCC" and "OOOOOOOO" will share only 1 bit in their fingerprints. And that lowers the similarity score to 0.08, much more accurate now.

mol = Chem.MolFromSmiles('CCCCCCCC')
fp1 = Chem.RDKFingerprint(mol, fpSize=64, nBitsPerHash=1)
print(fp1.ToBitString())

mol = Chem.MolFromSmiles('OOOOOOOO')
fp2 = Chem.RDKFingerprint(mol, fpSize=64, nBitsPerHash=1)
print(fp2.ToBitString())

print(DataStructs.FingerprintSimilarity(fp1, fp2))

# output 
0000000000000100000011000000100001000000000100000000000001000000
0000000000010000000000100100000001000000000000001000000000100010
                                 ^
0.07692307692307693