Folding atom pair count fingerprint

[Haonan-Zhang]

Hi,

I want to know how does rdkit fold a sparse atom pair count fingerprint into a fingerprint of given fpSize. I know that for the bit-vector fingerprint this is done by taking the modulo operation. But I find that this is not the case for the atom pair count fingerprint. To illustrate:

from rdkit.Chem import rdFingerprintGenerator
from rdkit import Chem

mol = Chem.MolFromSmiles('CCC')
fpgen = rdFingerprintGenerator.GetAtomPairGenerator(fpSize=2048)
fp = fpgen.GetCountFingerprint(mol)
print(fp.GetNonzeroElements())

{1311: 1, 1375: 2}

fp = fpgen.GetSparseCountFingerprint(mol)
print(fp.GetNonzeroElements())

{541730: 1, 558113: 2}

print(541730 % 2048)
print(558113 % 2048)

1058
1057
As you can see, the remainders are not the same. The problem is, I want to fold the BP fingerprint in the AtomPairs.Sheridan module (in the same way of AP fingerprint), which does not have the fpSize argument.

[greglandrum]

Hi @Haonan-Zhang, with the atom pair and topological torsion fingerprints it is possible to generate a bit position for an atom environment (one atom pair or one torsion) directly from the environment itself. This is what you get when you call GetSparseCountFingerprint() or GetSparseFingerprint(). These are hashed a second time and then "folded" (by doing modulo) when you call GetCountFingerprint() or GetFingerprint().

FYI, the Morgan fingerprints behave as you would expect: there is only one hashing step involved there:

In [7]: fpgen = rdFingerprintGenerator.GetMorganGenerator(fpSize=2048)

In [8]: fp = fpgen.GetCountFingerprint(mol)

In [9]: fp.GetNonzeroElements()
Out[9]: {80: 1, 294: 2, 1057: 2, 1344: 1}

In [10]: fp = fpgen.GetSparseCountFingerprint(mol)
    ...: [k%2048 for k in fp.GetNonzeroElements()]
Out[10]: [1344, 80, 1057, 294]

[Haonan-Zhang]

Hi Greg,
Thanks for your comment. Although I still don't get the hash function for the count fingerprint, I manage to create the hashed (folded) BP fingerprint by using the old fingerprint function:

from rdkit.Chem.rdMolDescriptors import GetHashedAtomPairFingerprint
from rdkit.Chem.AtomPairs import Sheridan

def typ_mol(mol):
    typMap = dict(CAT=1, ANI=2, POL=3, DON=4, ACC=5, HYD=6, OTH=7)
    typs = [typMap[x] for x in Sheridan.AssignPattyTypes(mol)]
    return typs

mol = Chem.MolFromSmiles('CCC')
ap = GetHashedAtomPairFingerprint(mol,2048, atomInvariants=typ_mol(mol))
print(ap.GetNonzeroElements())

I notice that GetAtomPairGenerator also has an atomInvariantsGenerator argument, but I don't know how to use it. So I will go with this walkaround for now.