1. PubChem
Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
PubChem Bioassays as a Source of
Introduction
Polypharmacology Methodology
Visualization
Application
Bin Chen, David Wild, Rajarshi Guha
School of Informatics
Indiana University
236th ACS National Meeting
17th August, 2008

2. PubChem
PubChem Bioassays Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Introduction
Currently contains 1157 assays Methodology
A number are follow ups of primary screens Visualization
Application
Assay size ranges from 2 to 224,000 molecules
Many compounds tested in multiple assays
PubChem web interface support queries that focus on
individual assays
Cross-assay queries can be tough

3. PubChem
Assay Content Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
AID 1 AID 2 AID 50
The data is obviously primary Introduction
Methodology
But the assay description and
Visualization
target are also useful pieces of
Application
information
Can we combine
data
AID 1 AID 2 AID 50
target
description
across multiple assays to draw
conclusions, gain insight?

4. PubChem
A Network Model of Bioassays - Goals Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
PubChem Guha
Bioassays
Introduction
PubMed PubChem
Methodology
Assay Network KEGG GO Visualization
Construction Application
Link into other data bases
Storage & Deployment
RDBMS
Network Interactive
Web Web Models Visualization
SQL
Service Page
Mapping to
External Networks
PPI Drug -Target

5. PubChem
Mapping Assay Networks to Real Networks Bioassays as a
Source of
Polypharmacology
Bin Chen, David
An assay network is an artificial network - does not Wild, Rajarshi
Guha
necessarily have physical meaning
We need to map the assay network onto a real biological Introduction
network Methodology
Visualization
PPI networks
Application
metabolic networks
drug target networks
Using the mapping, we’d like to identify MLSCN
compounds that might be active against one or more
nodes in the real network
The stepping stones . . .
How do we construct the assay network?
How do we map the network?

6. PubChem
Mapping Assay Networks to Real Networks Bioassays as a
Source of
Polypharmacology
Bin Chen, David
An assay network is an artificial network - does not Wild, Rajarshi
Guha
necessarily have physical meaning
We need to map the assay network onto a real biological Introduction
network Methodology
Visualization
PPI networks
Application
metabolic networks
drug target networks
Using the mapping, we’d like to identify MLSCN
compounds that might be active against one or more
nodes in the real network
The stepping stones . . .
How do we construct the assay network?
How do we map the network?

7. PubChem
Why Perform a Mapping? Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Identify compounds that interacts with two targets in Introduction
different pathways Methodology
Visualization
Alternatively, identify compounds that interact with a
Application
target in a pathway but not in another pathway
Identify compounds capable of disrupting protein-protein
interactions
Our ability to do these will depend on the quality of
assay data and the way we map the assay network to the
real network
Hopkins, A.L. et al, Curr. Opin. Chem. Biol, 2006, 16, 127–136

8. PubChem
Why Perform a Mapping? Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Identify compounds that interacts with two targets in Introduction
different pathways Methodology
Visualization
Alternatively, identify compounds that interact with a
Application
target in a pathway but not in another pathway
Identify compounds capable of disrupting protein-protein
interactions
Our ability to do these will depend on the quality of
assay data and the way we map the assay network to the
real network
Hopkins, A.L. et al, Curr. Opin. Chem. Biol, 2006, 16, 127–136

9. PubChem
Assay Network Construction Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Download Extract protein Guha
PubChem bioassay XML target ID's
Introduction
Methodology
Extract activity Evaluate
Visualization
scores pairwise
CLUSTAL Application
similarities
Exclude
compounds
with score
< 80
Connect assays if Connect assays if Connect assays if
they have their semantic their target
compounds in similarity is greater similarity is greater
common than X than X

10. PubChem
Assay Network Construction Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
We will focus on a compound-centric network Introduction
Methodology
A semantic network requires some form of annotation on
Visualization
the assays
Application
Initial attempts at annotation assays based on GO terms
(via descriptions)
Alternatively, could consider deriving annotations based
on the targets
Using protein target similarity restricts one to enzymatic
assays which leads to a relatively small assay network

11. PubChem
Assay Network Construction – Caveats Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Introduction
A compound-centric network is not very rigorous Methodology
Visualization
The PubChem activity score is known to be noisy
Application
Currently the only way to look at assay readouts over
the whole collection
Using an activity score cutoff of 80 is arbitrary
We haven’t considered promiscuity directly, though a
filter would be useful

12. PubChem
Assay Network - Common Compounds Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Introduction
Methodology
Visualization
Application

13. PubChem
Some Network Statistics Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Introduction
60
Methodology
222 assays with a single target
50
Visualization
Selected the smallest assay if
40
Application
Frequency
more than assay had the same
30
target
20
N = 125, E = 598
10
Vmax = 40, Vavg = 9.6
0
¯
C = 0.67 0 10 20 30 40
Vertex Degree
Histogram of vertex degree

14. PubChem
Clustering in the Assay Network Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Introduction
Methodology
Visualization
Application

15. PubChem
Assay Network - Common Compounds Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Introduction
Methodology
Visualization
Application

16. PubChem
Assay Network - Common Compounds Bioassays as a
Source of
Polypharmacology
388 targets NAD+ -dependent Bin Chen, David
Wild, Rajarshi
15-hydroxyprostaglandin Guha
dehydrogenase Introduction
Has active compounds common Methodology
with Visualization
pim-2-oncogene (505) Application
15-lipoxygenase (887)
aldo-keto reductase (381)
Luteonin Genistein

17. PubChem
Assay Network - Common Compounds Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Introduction
Methodology
Visualization
Application

18. PubChem
Assay Network - Common Compounds Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
749 and 755 target 5-HT1E and Introduction
5-HT1A respectively Methodology
Both have a (different) Visualization
compound in common with Application
1288 (selectin E)
Probably promiscuous given
that they are also active in
many other assays
But a selectin inhibitor is known
to reduce hyperalgesia by
blocking 5-HT3
Oliviera, M.C.G. et al, Neuroscience, 2007, 145, 708–714

19. PubChem
Assay Network - Common Compounds Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Most of these assay pairs have Introduction
Methodology
closely related targets
Visualization
Tissue non-specific alkaline Application
phosphatase and intestinal
alkaline phosphatase (1056 &
1017)
STAT1 and STAT3 (1303 &
1310)
ER-α and ER-β (1226 & 1228)
lethal factor (B. anthracis) and nF-κB (942 & 1309) have
one compound in common - podophyllotoxin

20. PubChem
Mapping an Assay Network Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Introduction
Methodology
Visualization
Application
Mapping
Function

21. PubChem
Defining a Mapping Function Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Introduction
Multiple mapping functions can be defined
Methodology
exact matches between assay target and external targets
Visualization
similarity between target sequences
Application
similarity between target binding sites
One could also map edges of one network onto another
Dependent on the nature of the external network
Depending on the nature of the definition, the mapping
procedure can be a trivial search or may require an
optimization scheme if multiple mappings are possible

22. PubChem
Assay Network to HPRD Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
The HPRD database collects protein-protein interaction Introduction
data and pathway membership Methodology
Visualization
The July 2007 release lists 31,708 PPI’s
Application
96 assays can be mapped to the unique proteins in
HPRD
We construct a HPRD network by identifying the pairs
from the 96 proteins that have a listed interaction
When mapping the HPRD network to the assay network,
we include singleton HPRD nodes

23. PubChem
HPRD Network Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Introduction
Methodology
Visualization
Application

24. PubChem
Assay - HPRD Network Mapping Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Introduction
Methodology
Visualization
Application

25. PubChem
Assay - HPRD Network Mapping Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Introduction
Methodology
Is this a useful mapping? Visualization
Application
Since we map assays to HPRD entries by target ID, we
aren’t getting new information on the assays individually
But we are able to easily identify assay targets that
interact with each other (or not)

26. PubChem
Comparing Two Assays Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Introduction
Methodology
Score = 95 Score = 100
Visualization
Application
CID 126298
AID 835 AID 1325
Gene: EPHB4 Gene: ABCB1
Signal Transduction Transporter
Axon guidance pathway ABC transporters
Over expressed in breast carcinoma Overexpression is related to multidrug
resistance in chemotherapy
Preferentially expressed in veins
Involved in the BBB
Required for angiogenesis
Not expressed very highly in vascular
tissue

27. PubChem
Comparing Two Assays Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
Introduction
Methodology
Score = 94 Score = 93 Visualization
Application
CID 647501
AID 903 AID 755
Gene: TP53 Gene: HTR1A
Nucleotide Metabolism Signal Transduction
Apoptosis, Cell Cycle, Cancer Neuroactive ligand-receptor
Controls cell cycle and apoptosis Target for anti-depressants
Inactivated in cancer cells

28. PubChem
Disrupting PPI’s Bioassays as a
Source of
Polypharmacology
Bin Chen, David
The pairs of interacting Wild, Rajarshi
Guha
targets have compounds
tested against both of them Introduction
Methodology
Majority are inactive or
Visualization
Tyrosine
3-monooxygenase
NF-kappa B inconclusive in both of them
activation protein Application
CID 1025314 is active in AID
445 but inactive in AID 903
Hypoxia Bcl-2
Inducible
P53
Factor

29. PubChem
Summary Bioassays as a
Source of
Polypharmacology
Bin Chen, David
Wild, Rajarshi
Guha
A network view of assays provides with a novel tool for Introduction
Methodology
visualization and summary of the assay collection
Visualization
It’s utility beyond visualization is dependent on the way Application
we construct the network
A compound-centric network allows us to use the assay
collection as a probe into external networks
Future work will investigate different forms of the assay
network focusing on protein target and GO annotation
similarity