Similarity

One way to rank and filter molecules from a larger set is to use similarity coefficients. One such way is by using Tanimoto. The experiment below runs similarity tests on a number of molecules using RDKit libraries (http://www.rdkit.org) and Python.

Installing RDKit

To install RDKit on Ubuntu 14.04 desktop OS, do:

sudo apt-get install python-rdkit librdkit1 rdkit-data

If you want to build and install RDKit from source, follow http://www.blopig.com/blog/2013/02/how-to-install-rdkit-on-ubuntu-12-04/.

For this experiment, we will use Zinc15 database (http://zinc15.docking.org/) to download a list of molecules in SMILES format, create their respective Morgan fingerprint as a sequence of 1s and 0s, and compare them to a hypothetical molecule to find similar ones using Tanimoto’s similarity.

The Code

The code listing given below is available on GitHub (https://github.com/jod75/chemoinformatics/blob/master/src/SimilarityTest.py).

#!/usr/bin/env python
 
###############################################################################
# Tanimoto Similarity test using RDKit and Zinc15 database
# Joseph D'Emanuele
#
 
import urllib
import os
from rdkit import Chem
from rdkit.Chem import AllChem
from rdkit import DataStructs
from rdkit.Chem import Draw
 
 
###############################################################################
# This runs through the molecules in a SMILES file and returns
# a list of Molecules.
def process_smiles_file(filename):
    smiles = {}
    with open(filename, "r") as infile:
        infile.readline()  # skip header
        for line in infile:
            parts = line.split()
            m = Chem.MolFromSmiles(parts[0])
            if m is None:
                continue
            smiles[parts[1]] = m
    return smiles
 
###############################################################################
# similarity test
 
###############################################################################
# create folders if necessary
if not os.path.exists("../data"):
    os.makedirs("../data")
if not os.path.exists("../out"):
    os.makedirs("../out")
 
# download smiles file
urllib.urlretrieve("http://files.docking.org/2D/AA/AAAA.smi", "../data/AAAA.smi")
 
# process file and create a list of Molecules
molecules = process_smiles_file("../data/AAAA.smi")
 
# use first molecule as query fingerprint
fp_query = AllChem.GetMorganFingerprintAsBitVect(molecules[molecules.keys()[0]], 2)
 
# dictionary to keep similarity index
similarities = {}
 
# compute Tanimoto similarity for all molecules in our file
for moleculeKey in molecules.keys():
    fp2 = AllChem.GetMorganFingerprintAsBitVect(molecules[moleculeKey], 2)
    similarity = DataStructs.FingerprintSimilarity(fp_query, fp2)
    similarities[moleculeKey] = similarity
 
# get top 20 similar molecules
top20 = sorted(similarities, key=similarities.get, reverse=True)[:20]
top20.insert(0, molecules.keys()[0])  # this is the query molecule
 
# get bottom 20 similar molecules
bottom20 = sorted(similarities, key=similarities.get, reverse=False)[:20]
bottom20.insert(0, molecules.keys()[0])  # this is the query molecule
 
# draw top20 similar molecules
img = Draw.MolsToGridImage([molecules[x] for x in top20], molsPerRow=2, subImgSize=(400, 400),
                           legends=["%s - %f" % (x, similarities[x]) for x in top20])
img.save("../out/similarities_top20.png")
 
# draw bottom20 similar molecules
img = Draw.MolsToGridImage([molecules[x] for x in bottom20], molsPerRow=2, subImgSize=(400, 400),
                           legends=["%s - %f" % (x, similarities[x]) for x in bottom20])
img.save("../out/similarities_bottom20.png")

Sample Output

The following images are sample output from the code above.

Query

ci_ts_query

Top matches

ci_ts_topmatch

No match

ci_ts_nomatch

Thanks for reading this post and happy screening.

Category: Similarity

Tanimoto Molecular Similarity Experiment