Transcript University of Cincinnati Compound Repository

Genomics Research Institute
University of Cincinnati
Compound Library
Wm. L. Seibel
January 10, 2007
Overview

Library Overview

Compound Characteristics
–
–
Design Concepts
Drug-Like

Library Screening Options

Summary of Library Advantages
Compound Repository

Haystack Neat Compound Storage
–
–
–

Solar (Solution Archive) – DMSO solutions
–
–


Capacity = 200,000 bottles
Current = 207,000 bottles
Freezer storage when appropriate
Capacity = 1.8 million tubes, 10,000 deep well (96) plates,
13,600 shallow well (384) plates
Current = 325,000 unique compounds
Related Compound Handling and Dissolution
instruments.
Housed at P&G’s Mason Business Center in ca. 3000
sf lab space
Haystack® Neat Chemical Storage
Haystack® Neat Chemical Storage
Haystack® Neat Chemical Storage
Solar® Solution Storage
Solar® Solution Storage
Solar® Solution Storage
Library Design Principles
Drug Lead Discovery
…greatly simplified
Target
Identification
Target
Validation
Compound
Screening
H2L Activity
Confirmation
Compound Selection
can greatly enhance
Efficiency
Compound
Optimization
LEAD
Library Compounds

The UC/GRI Compound Library is comprised of
compounds from four general categories:
–
–
–
–

1. Compounds purchased from numerous sources selected
to provide a diverse representation across “drug-like”
structural properties.
2. Compounds purchased that specifically target kinases
and GPCRs
3. Compounds prepared in-house specifically for projects in
kinases, GPCRs, phosphatases, ion channels and proteases
donated from P&G Pharmaceuticals.
4. Combinatorial Chemistry contract syntheses (Lower
Priority Cmpds).
This screening library is broadly diverse across druglike space, with enhanced concentrations in areas of
key biological relevance, including notably, kinases
and GPCRs.
P&G Pharma Selected Compounds

Chemically Diverse
–
–

Compounds selected based on drug like
properties (within “Drug-Like Space”)
–
–

Represented uniformly across drug-like space.
Want to ensure uniform, comprehensive and
diverse representation of compounds across the
structural & property types that are typical of drugs
and lead structures.
Chemical and Property Filters
Lipinski, Veber etc. rules
Total P&G investment to assemble repository
= $22 M (over past 10 years)
Chemical Property Filters
26 databases
>4 million structures
Vendor
Database
Remove duplicates
MW filter
Solubility Filter
Lipinski Rule of Five
•> 5 H-bond donors
•MW < 500
•c log P < 5
• N's + O's < 10
Remove reactives,
Unusual groups,
& toxicophores
(80 substructures)
“Cleaned”
database
Diversity Analysis
Describing Molecular Structure
Convert molecular structure into numerical values by
making computations of specific structural features
OH
Structure
Computations
N
Numeric
Descriptors
N
N
CH3
Relevant to
Binding Functions
Diversity Assessment Methodology

Used BCUT descriptors *
–
–
R.S. Pearlman, UT at Austin
DiverseSolutions (now available from Tripos)

Computed ~120 BCUTs

Selected a best subset of 6 BCUTs
–
6D space – visualization is a challenge
* J. Chem. Inf. Comput. Sci. 1999, 39, 11-20.
Pearlman’s BCUT descriptors *

6D Chem-space (structure-space)
– 2 atomic partial-charge descriptors
– 2 atomic polarizability descriptors
– a hydrogen-bond acceptor descriptor
– a hydrogen-bond donor descriptor
* http://www.awod.com/netsci/Issues/Jun96/feature1.html
Desc-2
Concept of Chemistry Space
Desc-1
Defining Drug Space
Based on structures of “drug-like” compounds from
 The World Drug Index (WDI)
 The Nation Cancer Institute Open Database
Desc-2
Desc-2

Desc-1
Desc-1
Diverse Subset Selection
Avoiding “redundant” representations
Diversity
Analysis
Compound Supply

External Suppliers (20+ vendors)
–
Brokerage Houses


–

Individual Compounds (Diversity)
Target Directed Libraries
Combinatorial Chemistry Companies
Corporate Suppliers
–
P&G Pharmaceuticals

Focus Areas - Medicinal Chemistry
Kinase, GPCR, Phosphatase, Ion Channel, proteases

Lead ID – Combinatorial Chemistry
Vendor “Dependability”
vendor
A
# tim es a
com pound
present in
# cm pds # com pounds # com pounds
Analytical
ave
ave
MS
evaluated
that w ere
that w ere not confirm ation conc purity confirm ed in
by MS
present in MS present in MS
rate
(m M) (%) an HTS assay
160
143
17
89.4%
1.9
95.9
0
B
171
142
29
83.0%
1.9
94.5
7
C
182
145
37
79.7%
6.4
94.0
34
D
31
23
8
74.2%
9.5
98.3
3
E
649
470
179
72.4%
5.8
93.1
117
F
1137
788
349
69.3%
5.1
91.1
426
G
47
32
15
68.1%
6.2
79.0
15
H
20
13
7
65.0%
4.4
93.9
5
I
230
0
0
63.3%
4.6
91.0
0
J
16
10
6
62.5%
6.1
91.7
7
K
258
152
106
58.9%
4.9
76.4
227
L
1906
962
944
50.5%
8.3
77.4
1803
M
651
266
385
40.9%
2.5
61.2
115
N
54
18
36
33.3%
3.9
87.0
14
O
48
15
33
31.3%
4.4
90.2
19
Totals:
Averages:
5560
3179
2151
2792
62.8%
5.1
87.7
On the Other Hand…

Even within Drug-like space, certain
classes can be somewhat clustered.

This library therefore has added
“focused libraries” from internal
synthesis and external vendors
emphasizing compounds relevant to:
–
–
GPCRs
Kinases
P&G Pharma Selected Compounds

Defined by experienced medicinal chemists
–

Broad, uniform distribution across Drug Space with
concentrations of density in key areas from directed
purchase and in house synthesis.
Compare to 5 vendor screening collections
–
–
–
3,000 to 500,000 compounds
27% - 56% of vendors’ compound collections do NOT
meet criteria for drug-like
UC Compound collection is 2X to 100X more
chemically diverse across Drug Space.
Vendor Libraries are inherently predisposed to clustered groupings.
We can pick the best, most relevant compounds from each.
Screening Library Options
Screening Library Design Options

Diverse broad collections
–
–

Class-associated compounds
–

Comprehensive screening against all available compounds
(ca. 250,000 cmpds)
Screening against a representative subset of available
compounds (e.g. 5000 cmpds)
Compounds with structural features often associated with a
particular target (e.g. kinases).
Structure-based compound selection
–
–
Virtual Screening of a crystal structure or high quality
homology model to identify the most likely inhibitors (ca.
2000 cmpds), followed by assay of these compounds.
Virtual Screening as above based on pharmacophore
models from known ligands of the target.
Diverse Subset Selection
Same Concept as Previously
250,000 Cmpd Library
5,000 Cmpd Abstract
Diversity
Analysis
Diverse Subset Selection
Execute Assay on subset of compounds
5,000 Cmpd Abstract
Identify Hits in Assay
MTS
Assay

Diverse Subset Selection
Pull Similar Compounds from original 250K Set
250,000 Cmpd Library

300 Cmpd Similarity Library
Similarity
Search

Diverse Subset Selection
Pull Similar Compounds from original 250K Set
300 Cmpd Similarity Library

Identify Hits in Assay
MTS
Assay




This Cycle can be repeated several times
until no new actives are found
Selection of
Nearest
Neighbors of Hits
Biological hit
Near neighbor
Iterative Cycling
5000 Cmpd
Representative
Library
~1000 Cmpd
NN Library
Assay
NN Search of
UC/GRI Library
~20 Cmpd
Hit List
2-3 Iterations
Final Set
NN Search of
Commercial
Compounds
~1000 Cmpd
NN Library
Assay
Final
Hit List
2-3 Iterations
~50 Cmpd
Hit List
Diverse Subset Selection
Pull Similar Compounds from Commercial 4.8M Set
300 Cmpd Similarity Library
4.8 M Commercial Library




Similarity
Search




Assay for actives, and cycle hits
back through similarity search loop.
Class-Associated Compounds
Select compounds similar to compounds known to intereact
with target class
250,000 Cmpd Library
15,000 Cmpd Library
Similarity
Analysis
Target Active
Virtual Screening
Screen GRI/UC library
Screen Commercial Cmpds
Iterative Cycling
5000 Cmpd
Representative
Library
~1000 Cmpd
NN Library
Assay
NN Search of
UC/GRI Library
~20 Cmpd
Hit List
2-3 Iterations
Final Set
NN Search of
Commercial
Compounds
~1000 Cmpd
NN Library
Assay
Hits of any origin
can enter the cycle
at this point.
Final
Hit List
2-3 Iterations
~50 Cmpd
Hit List
Hit to Lead Follow-up (H2L)
Hit to Lead Follow-up (H2L)

How to determine optimal hits for follow-up
–
–
–
Confirm ID and activity of hits
Cluster into groups of related compounds
Develop preliminary SAR info on each cluster

–
ID Key features for binding & selectivity
Assess Each Cluster for optmization







“Which compounds have fewest problems?”
Synthetic Ease
Proprietary Assessment
Selectivity Issues
Physical Properties
Metabolic Handles
Cellular Activity
Summary
UC/P&GP Library Advantages

Quality Advantages
–
–
–
–
–

Practical Advantages
–
–
–
–

Library carefully constructed to span drug-like space.
Compounds restricted to those with properties consistent with clinical materials.
Proven to produce viable hits for follow-up programs.
Comparisons have uniformly been favorable relative to commercial vendor sets.
Includes targeted subsets of compounds for key areas: GPCRs, Kinases,
Phosphatases, Ion channels.
SD file of structures and ID tags furnished for unrestricted use.
Many compounds from commercial sources, so resupply likely to be easy.
Materials supplied in microtiter plates (96 or 384) as requested.
Solution Stores made from local dry stores, so follow-up assays will be rapid.
Technical Advantages
–
–
Act as Liaison with screening group (internal or external).
Participate in advisory committee for compound acquisition decisions.
Library Use and Data Interpretation

Library Design Assistance
–
–
–
Computational assistance in selecting diverse subsets or
directed subsets.
Computational assistance in selecting compounds similar to
known leads (Nearest Neighbor).
Computational assistance in virtual screening by
pharmacophore or protein docking.
Library Use and Data Interpretation

Follow-up Assistance
–
–
–
–
–
–
Resupply assistance, synthesis info, supplier info
Assistance in obtaining related available compounds
(Similarity, substructure, Unity, Pharmacophore).
Provide preliminary lit search info (known info, IP, etc) on
prominent hits.
Clustering of hits into chemical/pharmacophore classes
included.
Provide help identifying chemistry groups with related
interests for collaborations
Provide assistance in connecting with contract chemistry
services (consult).
Questions
Thank you
Acknowledgements

Operations
–
–
Stacey Frazier
Kathy Gibboney

Management
–
–
–
–

Computational
–
–
–
Matt Wortman
David Stanton
Prakash Madhav
Ruben Papoian
Sandra Nelson
Joseph Gardner
Kenny Morand