Transcript PPT Slides - Center for Computational Sciences

Computational
Techniques in Support
of Drug Discovery
Jeffrey Wolbach, Ph. D.
October 2, 2002
Who Is Tripos?
Discovery Software
& Methods Research
Core Science &
Technology
Chemistry Products
& Services
Software
Consulting Services
Discovery Research
& Process Implementation
Sequential Drug Discovery
Choose
Disease


Target
Identification
Target
Validation
Lead
Identification
Lead
Validation
Lead
Optimization
ADME
Candidate
to Clinic
Many cycles of synthesis/testing to identify and optimize lead
Role of molecular modeling
o
o
o
o
unrealistic to jump from validated target to optimized lead
useful to reduce the number of synthesis/testing cycles
enables “first to file”
enlarge number of targets
Drug Discovery in Parallel
•
Choose a Disease
Target Identification
Target Validation
Knowledge-sharing environment:
genomics, HTS, chemistry, ADME,
toxicology
•
Collect more data, on more
compounds, more quickly
•
Apply predictive models of
“developability” early
Lead Identification
Lead Validation
•
•
Enhanced understanding &
predictive model building
Increase share of patented
time on market
Lead Optimization
ADME
Candidate to Clinic
Ligand-Based Design
Ligand Structures w/Activities
No Target Structure
Pharmacophore
Analysis
QSAR
Discern
Similarities and
Differences in
Active
Structures
Database Searching
New Candidate Structures for Synthesis/Testing
Pharmacophore Analysis
•
Assume active molecules share a binding mode
o
•
Don’t know binding mode, so active molecules are
considered flexible
o
o
•
Search for common chemical features of active molecules
Search set of pre-determined conformers
Allow molecules to flex during search
Typical features:
o
o
o
H-Bond Donors
H-bond acceptors
Hydrophobic groups
Pharmacophore Models
•
•
Chemical features in 3-D space
Distance constraints between chemical features
QSAR
•
•
Relates bioactivity differences to molecular structure
differences
Structure represented by numerical descriptors
o
o
•
Traditional (2D) QSAR
3D QSAR - CoMFA
Statistical techniques relate
descriptors to activity
+
D Activity
+
+
++
+
++
++
Activity = D0 + 0.5 D1 + 0.17 D2 + ...
+
+
D Descriptor
QSAR - Traditional (2D)
•
Descriptors are molecular properties
o
logP, dipole moment, connectivity indices ...
Structures + Activity
pKi=5.3
pKi=3.7
pKi=2.9
Descriptors
Predictive Model
(QSAR Equation)
logP = 1.9
m = 2.8
Estate = 7.2
pKi=A
+ B(logP)
+ C(m)
+ D(Estate)
+ ...
logP = 1.7
m = 2.3
Estate = 6.7
logP = 2.1
m = 3.5
Estate = 5.5
PLS
MLR
.
.
QSAR - 3D QSAR - CoMFA
•
•
•
Comparative Molecular Field Analysis
Descriptors are field strengths around molecules electrostatic, steric, H-bond ..
Fields can have easy physical interpretation
pKi=A + B(D1) + C(D2) + ...
QSAR/CoMFA - Interpretation
•
High Coefficient (important) lattice points can be
plotted around molecular structures
2D Database Searching
O
OH
O
O
N
S
N
O
N
O
•
010110010010101
O
Searches often performed on bit-strings
o
o
“Fingerprints” (many types)
Fingerprints display neighborhood behavior
•
Also includes substructure searching
•
Can search for similarity or dissimilarity
3D Database Searching
•
Query is a collection of features in 3-D space
o
o
•
Pharmacophore
Lead compound / specific atomic groups
Search a database
of flexible, 3-D
molecules
o
o
Molecules can’t
be stored in
every possible
conformation
Allow molecules
to flex to fit the
query
Example of
Structure-Based Design
3D Database Searching
•
•
Not restricted to
ligand-based
design
Information about
target can be
included in the
query
o
o
o
Can define steric
hindrances
Additional
interaction sites
Serves to filter
hits from the
search
Identification of Novel Matrix
Metalloproteinase (MMP) Inhibitors
MMPs
•Zinc-dependent proteases
•Involved in the degradation and
remodeling of the extracellular
matrix
They are important
therapeutic targets with
indications in:
A fibroblast collagenase-1 complexed with a diphenylether sulphone-based hydroxamic acid
•Cancer
•Arthritis
•Autoimmunity
•Cardiovascular disease
Objectives
Design high affinity MMP inhibitors based on the
diketopiperazine scaffold by:
•Creating a virtual combinatorial library of candidate inhibitors
•Using virtual screening tools to identify candidates with the
highest predicted affinity
•Perform R-group and binding mode analysis to guide library
design
Synthesis of
DKP-MMP inhibitors
DKP-I
DKP-II
1.) Esterification of the solid support (HO-) with an amino acid
2.) Reductive alkylation of the amino acid and acylation of the resulting
secondary amine
3.) Deprotection of the N-alkylated dimer followed by cyclic cleavage from the
resin yielding diketopiperazine (DKP)
Finding & Filtering Reagents
UNITY 2D structure search of the ACD
Filtered out:
•Metals
•MW > 250
•RB > 8
Filtered out:
•Metals
•MW > 400
•RB > 15
1154 aldehydes
73 Boc protected amino acids
Selecting Reagents &
Building the Virtual Library
Selector™
Legion™
Diverse selection of
amino acids (R1) and
aldehydes (R2) using:
Model the reaction and create
virtual combinatorial library
•2D Finger Prints
•Atom Pairs
•Hierarchical Clustering
Randomly selected 14
amino acids for R3
55 amino acids
x 95 aldehydes
x 14 amino acids
= 73,150 compounds
(R1)
(R2)
(R3)
(~75k, 8.5 MB)
The CombiFlexX Protocol
•Select a diverse subset of compounds using OptiSim
•Dock and score the compounds in the diverse subset
using FlexX
•Select unique core placements using OptiSim
•Hold each core placement fixed in the binding site as
each R-group is independently attached, docked, and
scored.
•Sum the scores of the "R-cores" and subtract the
score of the common core
Computation times scale as the sum of the number of Rgroups rather than as the product of the number of R-groups
Virtual Screening
of DKP-MMP Inhibitors
• ~75k compound library
• MMP target structure
collagenase-1
(966c.pdb)
• 150 diverse compounds
selected and docked
• 39 non-redundant core
placements based on
RMSD > 1.5 Å.
Virtual Screening Results
•Docked 91% of the library
•36 compounds/minute
•331 compounds predicted
to be more active than
those published
Consensus Scoring Results
•Extracted the top 1000 library
compounds based on Flex-X score
•Ranked the top library compounds
and published “highly actives”
using CScore
•10 compounds predicted by all
scoring functions to be more active
than “highly actives”
R-Group Analysis in
HiVol
Frequency of R-group use among 331 active virtual compounds
R1
(55 reagents)
R2
(95 reagents)
R3
(14 reagents)
R-Group Analysis in
HiVol, con’t.
Frequency of R-group use among 331 active virtual compounds
R1
Summary
Used CombiFlexX and HiVol to:
• Identify highly promising candidates
• Perform R-group analysis & Binding mode analysis to guide further
computational design of libraries
Further Work
• Diversity/similarity analysis of the published and virtual libraries
• Use docking results for library design in Diverse Solutions
• SAR development
Binding Mode Analysis
Frequency of core placement use
R-Group Analysis in
HiVol
Frequency of R-group use among 331 active virtual compounds
R1
(55 reagents)
R2
(95 reagents)
R3
(14 reagents)