Substructure filters for hit triaging
Hit triaging code reported from Novartis Institutes for BioMedical Research (NIBR).
Introduction
Now there are lots of substructure filters for getting drug like molecules such as PAINS, LillyMedChem rules etc. And most of rules are publically available. It's reall great thing.
And recently new filter is disclosed from NIBR. And the rules are provided from rdkit/Contrib/NIBRSubstructureFilters ;) The LinkURL is below. https://github.com/rdkit/rdkit/tree/master/Contrib/NIBRSubstructureFilters
The orignal code is command line tool for batch analysis so it is useful for handling SMILES set as CSV format. I'm interested in the code and tried to use it from jupyter notebook.
import os
import sys
import pandas as pd
import numpy as np
from rdkit import Chem
nirb_dir = os.path.join('/home/iwatobipen/miniconda3/envs/chemoinfo/share/RDKit/Contrib/NIBRSubstructureFilters/')
sys.path.append(nirb_dir)
from collections import Counter, defaultdict, namedtuple
from rdkit import Chem
from rdkit.Chem import FilterCatalog
from rdkit.Chem import rdMolDescriptors
from rdkit.Chem.Draw import IPythonConsole
from rdkit.Chem import Draw
import assignSubstructureFilters
FilterMatch = namedtuple('FilterMatch', ('SubstructureMatches', 'Min_N_O_filter', 'Frac_N_O', 'Covalent', 'SpecialMol', 'SeverityScore'))
# Build the filter catalog using the RDKit filterCatalog module
def buildFilterCatalog():
inhousefilter = pd.read_csv(os.path.join(nirb_dir,'SubstructureFilter_HitTriaging_wPubChemExamples.csv'))
inhouseFiltersCat = FilterCatalog.FilterCatalog()
for i in range(inhousefilter.shape[0]):
mincount=1
if inhousefilter['MIN_COUNT'][i] != 0:
mincount = int(inhousefilter['MIN_COUNT'][i])
pname = inhousefilter['PATTERN_NAME'][i]
sname = inhousefilter['SET_NAME'][i]
pname_final='{0}_min({1})__{2}__{3}__{4}'.format(pname,mincount,
inhousefilter['SEVERITY_SCORE'][i],
inhousefilter['COVALENT'][i],
inhousefilter['SPECIAL_MOL'][i])
fil = FilterCatalog.SmartsMatcher(pname_final,inhousefilter['SMARTS'][i], mincount)
inhouseFiltersCat.AddEntry(FilterCatalog.FilterCatalogEntry(pname_final,fil))
inhouseFiltersCat.GetEntry(i).SetProp('Scope', sname)
return inhouseFiltersCat
def assignFilters(mol):
results=[]
inhouseFiltersCat = buildFilterCatalog()
NO_filter = '[#7,#8]'
sma = Chem.MolFromSmarts(NO_filter, mergeHs=True)
qc,NO_filter,fracNO,co,sc,sm = [np.NaN]*6
# fraction of N and O atoms
numHeavyAtoms = mol.GetNumHeavyAtoms()
numNO = len(mol.GetSubstructMatches(Chem.MolFromSmarts('[#7,#8]')))
fracNO = float(numNO)/numHeavyAtoms
# all substructure filters
entries = inhouseFiltersCat.GetMatches(mol)
if len(list(entries)):
# initialize empty lists
fs,sev,cov,spm = ([] for _ in range(4))
# get the matches
for entry in entries:
pname=entry.GetDescription()
n, s, c, m = pname.split('__')
fs.append(entry.GetProp("Scope")+'_'+n)
sev.append(int(s))
cov.append(int(c))
spm.append(int(m))
# concatenate all matching filters
qc = ' | '.join(fs)
# assign overall severity
if sev.count(2):
sc = 10
else:
sc = sum(sev)
# get number of covalent flags and special molecule flags
co = sum(cov)
sm = sum(spm)
# if non of the filters matches
else:
qc = 'no match'
sc = 0
co = 0
sm = 0
# special NO filter
if not mol.HasSubstructMatch(sma):
NO_filter = 'no_oxygen_or_nitrogen'
else:
NO_filter = 'no match'
res = FilterMatch(qc,NO_filter,fracNO,co,sm,sc)
return res
Now almost there. Let's check the code. Following molecules are came from original repo.
m1 = Chem.MolFromSmiles('Fc1ccc(c(Cl)c1Cl)n2nnnc2Cc3cccnc3')
m2 = Chem.MolFromSmiles('[O-][N+](=O)c1ccc(cc1)C(=O)Oc2cccc3oc(cc23)c4ccccc4')
Draw.MolsToGridImage([m1, m2])
res1 = assignFilters(m1)
res2 = assignFilters(m2)
print(res1)
print(res2)
As you can see, running the code on jupyter, we can get molecule with highlight atoms.
m1
m2
