The document discusses issues with data quality in public domain databases used for drug repurposing, noting errors that proliferate between databases as data is shared and sourced. It advocates for collaboration on data curation efforts, adopting standards for data representation and licensing, and developing apps and semantic web approaches to facilitate crowdsourcing data analysis and feedback. The goal is to improve data quality to enable more accurate computational modeling for drug discovery.
1. http://tinyurl.com/d6wodsl
Mining public domain data as a basis
for drug repurposing
Antony J Williams, Sean Ekins and Valery Tkachenko
ACS Philadelphia August 2012
2. Drug Repurposing
Drug repurposing commonly
means data reexamination also!
Lots of data mining occurs
Then more screening which
creates more data..
LOTS of public databases used
to examine repurposing…
5. Where do you get your data?
Databases?
Patents?
Papers?
Your own lab?
Collaborators?
All of the above?
What is likely common to all sources? Data
Quality issues. There is no perfect database.
6. Public Domain Databases
Our databases are a mess…
Non-curated databases are proliferating errors
We source and deposit data between databases
Original sources of errors hard to determine
Curation is time-consuming and challenging
7.
8. Availability of libraries of FDA drugs
Johns Hopkins Clinical Compound library- made compounds available at cost
17. NCATS Discovering “New Therapeutic
Uses for Existing Molecules”
58 Molecule names
and identifiers. Where
are the “structures”?
18. NCATS dataset
• Several groups tried to collate molecules
• Chris Lipinski provided approximately 30 unique molecules
• Simple molecule descriptors shows no difference between
compounds classified as discontinued (N= 15) or those in
clinical trials (n = 14).
• Where is the definitive set of publicly accessible molecules
for computational repurposing and analysis?
19. Drug structure quality is important..
Many groups ARE doing in silico repositioning
Integrating or using sets of FDA drugs..and if
structures are incorrect predictions will be
Where is the definitive set of FDA approved
drugs with correct structures?
Ideally we need linkage between in vitro data
and clinical data
20. We have a problem…
Lots of data available but quality is suspect
Errors proliferate database to database
Data continues to flow in unabated
When errors are identified hard to get fixed!
Data licensing is confusing – “Open Data”
We are “takers” not “givers” mostly…
Standards are lacking:
Data licensing
Data processing – structure standardization
21. So what needs to happen to improve?
• Let’s agree collaboration and crowdsourcing
can help
• Provide SIMPLE ways to provide feedback
• Contribute when possible – databases should
provide feedback mechanisms
• Adopt standards for structure handling and
representation
• Adopt standards for data interchange
• Allow machine handling of data – use the
power of the semantic web
26. “Appify” curation and collaboration
• The data network is complex
• “Appify” collaboration and
curation networks
• Increasing crowdsourcing role
for data analysis
Ekins & Williams, Pharm Res, 27: 393-395, 2010.
28. Open Drug Discovery Teams
Free iOS app used to expose repurposing data
All of this data has been tweeted
http://tinyurl.com/6l9qy4f
Ekins, Clark and Williams, Mol Informatics, in Press 2012
30. Simple Rules for licensing “open” data
Gather stakeholders. Decide if goals are primarily scientific,
commercial or mixed.
Explore benefits of open licensing and drawbacks of
enclosure. Hold closely to open definitions and standards.
Do not write your own IP licenses!
Provide simple explanations for terms of use. Use
metadata to indicate licensing terms explicitly - the
Creative Commons Rights Expression Language is a
good tool.
Do not lock up metadata. If you can’t make the data public
domain, make the metadata public domain.
Williams, Wilbanks and Ekins.
PLoS Comput. Biol. in Press Sept.2012
31. Open PHACTS Project
Develop a set of robust standards…
Implement the standards in a semantic integration hub
Deliver services to support drug discovery programs
in pharma and public domain
22 partners, 8 pharmaceutical companies, 3 biotechs
36 months project
Guiding principle is open access, open usage, open source
- Key to standards adoption -
32.
33.
34. To facilitate THIS process!
IP?
What’s the
structure?
Are they in
our file?
What’s
similar?
What’s the
Pharmacology target?
data?
Known
Pathways?
Competitors?
Working On
Connections Now?
to disease?
Expressed in
right cell type?
36. Taxol: Paclitaxel Bioassay Data
Most Bioassay data associated with structure
with one ambiguous stereocenter
37. Measuring data: dispensing dependencies
Data from 2 AstraZeneca patents - Ephrin pharmacophores
developed using data for 14 compounds with IC50. Different
dispensing methods give different results. Impact
hypotheses and could impact drug discovery.
Acoustic Disposable tip
Hydrophobic Hydrogen Hydrogen Observed vs.
features (HPF) bond acceptor bond donor predicted IC50
(HBA) (HBD) r
Acoustic mediated process
2 1 1 0.92
Disposable tip mediated process
0 2 1 0.80
Ekins, Olechno and Williams, Submitted 2012
38. Measuring data: dispensing dependencies
Acoustically-derived IC50 values were 1.5 to 276.5-fold
lower than for tip-based dispensing
• Pharmacophores and other computational models are used
to guide medicinal chemistry.
• Non tip-based methods may improve HTS results and avoid
misleading computational and statistical models.
• No analysis of influence of dispensing processes on data.
• Public databases should annotate metadata to create larger
datasets for comparing different computational methods.
How much data is reproducible, accurate, valid? The
challenge of high-throughput science.