MHFP

MHFP6 (MinHash fingerprint, up to six bonds) is a molecular fingerprint which encodes detailed substructures using the extended connectivity principle of ECFP in a fundamentally different manner, increasing the performance of exact nearest neighbor searches in benchmarking studies and enabling the application of locality sensitive hashing (LSH) approximate nearest neighbor search algorithms. To describe a molecule, MHFP6 extracts the SMILES of all circular substructures around each atom up to a diameter of six bonds and applies the MinHash method to the resulting set. MHFP6 outperforms ECFP4 in benchmarking analog recovery studies. Furthermore, MHFP6 outperforms ECFP4 in approximate nearest neighbor searches by two orders of magnitude in terms of speed, while decreasing the error rate.

Associated Publication: https://jcheminf.biomedcentral.com/articles/10.1186/s13321-018-0321-8

Materials

The performance of MHFP has been avaluated using Open-source platform to benchmark fingerprints for ligand-based virtual screening and ChEMBL24. Python (pyspark) scripts for the ChEMBL24 benchmark can be found in this repository.

Dependencies

MHFP requires Python 3.x and Numpy.

MHFP requires the cheminformatics library RDKit.

Installation

MHFP can be installed using the Python packet manager pip:

pip install mhfp

Usage

Usage is straightforward:

from mhfp.encoder import MHFPEncoder

mhfp_encoder = MHFPEncoder()

fp_a = mhfp_encoder.encode('CCOC1=C(C=C(C=C1)S(=O)(=O)N(C)C)C2=NC(=O)C3=C(N2)C(=NN3C)C(C)(C)C')
fp_b = mhfp_encoder.encode('CCCC1=NN(C2=C1NC(=NC2=O)C3=C(C=CC(=C3)S(=O)(=O)N4CCN(CC4)C)OCC)C')
fp_c = mhfp_encoder.encode('O=C(OC)C(C1CCCCN1)C2=CC=CC=C2')

print(MHFPEncoder.distance(fp_a, fp_a))
print(MHFPEncoder.distance(fp_a, fp_b))
print(MHFPEncoder.distance(fp_a, fp_c))
print(MHFPEncoder.distance(fp_b, fp_c))

#> 0.0
#> 0.45849609375
#> 0.97998046875
#> 0.97216796875

LSH Forest

As the fingerprints are created using the locality sensitive hashing scheme MinHash, they are ready to be used with LSH-based algorithms:

from mhfp.encoder import MHFPEncoder
from mhfp.lsh_forest import LSHForestHelper

mhfp_encoder = MHFPEncoder()

fp_a = mhfp_encoder.encode('CCOC1=C(C=C(C=C1)S(=O)(=O)N(C)C)C2=NC(=O)C3=C(N2)C(=NN3C)C(C)(C)C')
fp_b = mhfp_encoder.encode('CCCC1=NN(C2=C1NC(=NC2=O)C3=C(C=CC(=C3)S(=O)(=O)N4CCN(CC4)C)OCC)C')
fp_c = mhfp_encoder.encode('O=C(OC)C(C1CCCCN1)C2=CC=CC=C2')
fp_q = mhfp_encoder.encode('COC1=C(C=C(C=C1)S(=O)(O)N(C)C)C2=NC(=O)C3=C(N2)C(=NN3C)C(C)(C)C')

lsh_forest_helper = LSHForestHelper()

fps = [fp_a, fp_b, fp_c]

for i, fp in enumerate(fps):
    lsh_forest_helper.add(i, fp)

lsh_forest_helper.index()

print(lsh_forest_helper.query(fp_q, 1, fps))

#> [0]

SECFP

SECFP (SMILES Extended Connectifity Fingerprint) creation is achieved by calling the static method MHFPEncoder.secfp_from_smiles():

from mhfp.encoder import MHFPEncoder
fp = MHFPEncoder.secfp_from_smiles('CCOC1=C(C=C(C=C1)S(=O)(=O)N(C)C)C2=NC(=O)C3=C(N2)C(=NN3C)C(C)(C)C')

Documentation

MHFPEncoder

The class for encoding SMILES and RDKit molecule instances as MHFP fingerprints.

__init__(n_permutations = 2048, seed = 42)

MHFPEncoder.encode(in_smiles, radius = 3, rings = True, kekulize = True, sanitize = True)

Returns a numpy.ndarray containing the fingerprint hashes.

MHFPEncoder.encode_mol(in_mol, radius = 3, rings = True, kekulize = True)

Returns a numpy.ndarray containing the fingerprint hashes.

MHFPEncoder.distance(a, b)

Returns a float representing the distance between two MHFP encoded molecules.

MHFPEncoder.secfp_from_mol(in_mol, length=2048, radius=3, rings=True, kekulize=True)

Returns a numpy.ndarray containing the fingerprint values.

MHFPEncoder.secfp_from_smiles(in_smiles, length=2048, radius=3, rings=True, kekulize=True, sanitize=False)

Returns a numpy.ndarray containing the fingerprint values.

LSHForestHelper

If you know your way around Python, I suggest you use the LSHForest class directly. See code for details.

__init__(dims = 2048, n_prefix_trees = 64)

LSHForestHelper.add(key, mhfp):

LSHForestHelper.index():

Has to be run after adding entities and before running queries!

MHFPEncoder.query(query_mhfp, k, data, kc=10)