pdf 609 K

MCSS in OEChem
Krisztina Boda, Matthew Stahl
OpenEye CUP IX, Santa Fe, March 2008

Overview
• What is MCSS?
• Why chemists might need MCSS?
• Defining MCSS in OEChem
– How to define atom and bond equivalency?
– How to evaluate detected common substructures?
– How to traverse the search tree
• Example code

Maximum Common Substructure
The largest sub-graph shared by two molecular structures.
N
m!
n!
( m − k)(
n − k)!
k!
O
O
Cl
NP-complete problem
n, m = number of nodes in the compared graphs
k = number of nodes in the common sub-graph
Cl
O
O

Possible Applications of MCSS
• clustering compound libraries
• diversity measures of a compound
library
• mapping chemical reactions
• R-group decomposition

Atom/Bond Equivalency
OEExprOpts::DefaultAtoms =
AtomicNumber|Aromaticity|FormalCharge
Cl
O
N
Cl
O
Cl
N
N
O
N
N
F
Cl
O
O
Cl
OEExprOpts::DefaultBonds =
BondOrder|Aromaticity
O
N
N
N
N
Cl
F
O
O
N
N
N
F

OEExprOpts Examples
DefaultAtom|EqAromatic
DefaultBond
DefaultAtom|EqHalides
DefaultBond
DefaultAtom
DefaultBond|EqSingleDouble
Cl
Cl
Cl
O
N
O
N
O
N
Cl
Cl
Cl
O
O
O
O
O
N
N
N
N
O N
N
N
N
N
F
F
F

MCS Scoring Functions
OEChem built-in scoring functions
– OEMCSMaxAtoms [default]
– OEMCSMaxBonds
– OEMCSMaxAtomsCompleteCycle
– OEMCSMaxBondsCompleteCycle
Ability to use user-defined scoring function

OEMCSMaxBonds
scoring function
NB +
NA
100
NB = number of bonds
NA = number of atoms
S
S
S
N
N
N
S
Br
S N
Br
S N
Br
N
score
5.06
5.06
5.05

OEMCSMaxBondsCompleteCycle
scoring function
NA
NB P*
NUCB
100 − +
NB
NA
P
NUCB
= number of bonds
= number of atoms
= penalty (default 1.0)
= number of unmapped
cyclic query bonds
S
N
S
Br
N

Unique/Non-Unique MCSS
An MCSS match is
considered unique:
• if it differs from all other
matches found previously by
at least one atom or bond
• if it represents different
mapping
O
O
O
O
Example for unique maximum common
substructures
O
O

Exhaustive MCSS
Pro:
- guarantee to find the
maximum common
substructure(s)
Contra:
- complex structures
cannot be mapped
efficiently
N
for each matching
atom pair
…
O
O
Cl
…
…
N
N
N O
Cl
O
O
O
O
?
O
O
Cl
Cl
? ?
O
Cl
Cl
common sub-graph
can not be extended
any more
O
O
Cl
?
Cl
evaluate
O
O
O
O
…
extend
subgraph
all
possible
ways

Approximate MCSS
pre-defined paths in the pattern
N O
…
…
O
N O
O
try to follow the same path in
the target
Cl
Cl
_
Cl
Cl
O
O
O
O

Comparison _ Speed
on average the approximate search 3-6 times faster
O
O
O
O
O
H
H
approximate
0.4 sec (CPU)
3 matches
O
O
O
H
O
O
O
O
O
O
O
O
O
O
H
H
O
O
O
O
O
O
O
O
O
O
exhaustive
100.1 sec (CPU)
59 matches
O
H
H
H
–
O
O

Exhaustive & Approximate MCSS
S
S
N
N
S
S
N
N
O
O
approximate
exhaustive
N
N
O
O
S
S
O
O
N
N

Comparison _ Accuracy
3000 compounds
from Maybridge
9M comparisons
OEMCSMaxAtoms()
approximate 3_
faster

MCSS Example
#!/usr/bin/env python
from openeye.oechem import *
import os,sys
pattern = OEGraphMol()
target = OEGraphMol()
OEParseSmiles(pattern, "c1cc(O)c(O)cc1CCN")
OEParseSmiles(target, "c1c(O)c(O)c(Cl)cc1CCCBr")
atomexpr = OEExprOpts_DefaultAtoms
bondexpr = OEExprOpts_DefaultBonds
mcstype = OEMCSType_Exhaustive
mcss = OEMCSSearch(pattern,atomexpr,bondexpr,mcstype)
mcss.SetMCSFunc(OEMCSMaxAtoms())
mcss.SetMinAtoms(6)
unique = True
count = 1
for match in mcss.Match(target,unique):
sys.stdout.write("\nMatch %d :" % count)
sys.stdout.write("\npattern atoms: ")
for ma in match.GetAtoms():
sys.stdout.write("%d " % ma.pattern.GetIdx())
sys.stdout.write("\ntarget atoms: ")
for ma in match.GetAtoms():
sys.stdout.write("%d " % ma.target.GetIdx())
count+=1
m = OEGraphMol()
OESubsetMol(m,match,True)
smi = OECreateCanSmiString(m)
sys.stdout.write("\nmatch smiles = %s \n" % smi)
10
12
Cl
N
11
9
10
8
9
Cl
Match 1:
pattern atoms: 0 1 2 3 4 5 6 7 8 9
target atoms: 7 5 3 4 1 2 0 8 9 10
match smiles = CCc1ccc(c(c1)O)O
7
0
8
7
6
1
0
5
O
4 5
2 3
6
O
O
1 2
3 4
O

Conclusion
• Adaptable MCSS
– define atom/bond equivalency
– set scoring function
– choose search method
• The approximate search provides a
good trade-off between accuracy and
speed

Future Development
Automatic R-group decomposition
Cl
R 1
Cl
N
R 3
O
O
N
core
R 2
O
O
R 1
R 2
R 3
N
Cl
O
O
Cl
O
O
O
O
N +
O –
N
O
…
H …
N +
O –
O
…
…

Link To Resources
• http://www.eyesopen.com/docs/
– OEChem – API (html/pdf)
– OEChem – C++ Theory Manual (html/pdf)
• Example
Chapter 17. Pattern Matching