Presentation to the EPA (August 2016) about the BioAssay Express project, from Collaborative Drug Discovery. Describes the history and potential of the project, with the intention of opening a dialog about incorporating EPA toxicity data.

1. Machine-readable assay protocols
Alex M. Clark
http://collaborativedrug.com

2. The 21st century (supposedly)
2
504854 449764
A Cell Based Secondary Assay To Explore Vero
Cell Cytotoxicity of Purified and Synthesized
Compounds that Inhibit Mycobacterium
Tuberculosis (4)
This functional assay was developed for
detection of compounds inhibiting Vero E6 cells
viability as a secondary screen to the beta-
lactam sensitizing M. tuberculosis bacteriocidal
assay.
In this assay, we treated Vero E6 cells with
compounds selected as hits in the M.
tuberculosis assay for 72 hours over a 10 point
2-fold dilution series, ranging from 0.195 uM to
100 uM. Following 72 hours of treatment,
relative viable cell number was determined
using Cell Titer Glo from Promega. Each plate
contained 64 replicates of vehicle treated cells
which served as negative controls.
Outcome: Compounds that showed <70% cell
viability for at least one concentration were
defined as "Active". If the % viability at all doses
was >70%, the compound was defined as
"Inactive".
...
A High Throughput Confirmatory Assay used to
Identify Novel Compounds that Inhibit
Mycobacterium Tuberculosis in the absence of
Glycerol
Outcome: Compounds that showed >30% inhibition
for at least one concentration in the Mtb dose
response were defined as "Active". If the inhibition
at all doses was <30% in the Mtb assay, the
compound was defined as "Inactive". In the primary
screen a compound was deemed "Inactive" if it had
a Percent Inhibition <70.31%. Compounds with a
Percent Inhibition >70.31% but were not selected
for follow up dose response were labeled
"Inconclusive."
The following tiered system has been implemented
at Southern Research Institute for use with the
PubChem Score. Compounds in the primary screen
are scored on a scale of 0-40 based on inhibitory
activity where a score of 40 corresponds to 100%
inhibition. In the confirmatory dose response
screen, active compounds were scored on a scale
of 41-80 based on the IC50 result in the Mtb assay
while compounds that did not confirm as actives
were given the score 0.
...
69%
similar (?)

3. The 21st century (actually)
3
69%
similar

4. Common Assay Template
• Built from semantic web ontologies:
BioAssay Ontology (BAO)
Drug Target Ontology (DTO)
Cell Line Ontology (CLO)
Gene Ontology (GO)
... and others
• Each annotated term is a URI: compatible with the
universe of linked data
4

5. Schema Editor
• Templates can be customised:
• Categories serve as guidance, and UI design
5

6. New Templates
• Editor and schema are open source:
http://github.com/cdd/bioassay-template
• The Common Assay Template (CAT) designed
mainly for assays from NIH Molecular Libraries
Program: to capture high level characteristics
• Personalised templates can be created
• Other general templates are planned: Toxicity ?
6

7. Content Creation
• Refining the template & user interface, and generating
training data, using text from PubChem assays:
7

8. Machine Learning
• Existing data used for learning: proposed annotations
8
natural
language
processing
Bayesian
models

9. Ontology Branches
• Each annotation is part of a
tree structure
• Inheritance provides extra
layers of meaning
• From common assay
template + ontologies
• Updated as ontologies are
extended or modified
9

10. Searching
• Design a search
using semantic
terms
• Similar interface
as annotation
10

11. Results
• Ranked by using annotations like fingerprints...
11

12. Browsing
• Select primary category
• Toggle desired terms
• Select secondary
category (etc)
12

13. Interoperability
• Service auto-downloads assays from PubChem:
whitelisted sources (mainly MLPCN)
• Curated assays can be downloaded via API
using semantic web formats (Turtle/RDF/JSON-LD)
• Adding finalised content currently by manual upload
• Discussions with PubChem regarding two way
communication, e.g. sending annotations back to the
original assay record
13

14. Deployment
• www.bioassayexpress.com is public
currently used as a read-only service
evolving rapidly with new experimental features
free as in beer, not speech
• Private installations and derived products are planned
• Key features will be integrated into CDD Vault & ELN
• Core components & data model open source
(GitHub): community adoption is welcomed
14

15. Probes Report
• Possible use case
(work in progress)
• Rows: NIH probe
compounds
• Columns: Curated
assay targets
• Retroactive look at
the data: more of it,
better quality
15

16. Future Features
• Semantic annotation of assays: make it really easy
ELNs are immediate benefactors
tag, organise, search for assay protocols
• Model building: thousands of datasets - which of them
are measuring the same property in the same way?
• Deep machine learning analysis (open ended!)
• Advanced schema: encode the entire protocol
• An alternative to text: annotate first, automatic report
16

17. Toxicity Data
• One approach might be to design a new schema
(e.g. the Toxicity Assay Template)
• Submission of data such as ToxCast into PubChem,
divided into assays and compounds...
• ... trigger automatic downloading into BioAssay
Express, allowing them to be annotated
• Annotating a large number of toxicity assays would
open a universe of possibilities
17

18. Acknowledgments
• Biologists
Janice Kranz
Haifa Ghandour
Karen Featherstone
18
• Collaborative Drug Discovery
Barry Bunin
Kellan Gregory
and the rest of the team
• More information
http://github.com/cdd/bioassay-template
http://www.bioassayexpress.com
http://collaborativedrug.com
PeerJ
Comp Sci
2:e61
(2016)