This document discusses Cubist Pharmaceuticals' DT workbench, a web-based platform for integrating drug design tools. It introduces drug metabolism and how computational models can help predict sites of metabolism to improve drug candidates. The workbench standardizes inputs using templates and utilizes Marvin components for drawing, viewing, and analyzing structures. A case study demonstrates using the workbench to identify the metabolic liability of Ticrynafen, perform lead hopping to find alternatives, and evaluate new structures. The workbench reduces learning curves and allows customizing Marvin tools for various needs like highlighting predicted sites of metabolism.

1. The Shape of Cures to Come™
Cubist Pharmaceuticals
A chemistry friendly system integrating drug
design tools and a consistent visual interface
Xin Zhang, Christian Baber

2. Outline
 Introduction to drug metabolism
 Introduction to Cubist Discovery Informatics group
 The DT workbench
 Case study

3. Outline
 Introduction to drug metabolism
 Introduction to Cubist Discovery Informatics group
 The DT workbench
 Case study

4. Drug metabolism
 Convert drugs into other products (metabolites) within the body
 Controls the duration and intensity of the action of many drugs
 Over 50% of drug development failures
can be associated with metabolic effects
 In general metabolites are more easily
excreted than the original drug
 Secondary pharmacological implications
– Normally pharmacologically inert
– Sometime they are pharmacologically active.
• similar activity to the drug
• or a different activity
• or be toxic

5. Ticrynafen
 A diuretic drug to treat hypertension
 Withdrawn in 1980 only a few months after launched
 Several hundred cases of liver injury reported, 25 deaths
 Covalent binding of metabolites to human liver microsomes reported
Cytochrome P450 isozyme 2C9
(CYP2C9)
Thiophene epoxide Thiophene sulfoxide
Reactive
Reactive

6. RS Predictor – A SOM Prediction Model
 Models predict site of metabolism (SOM) for given structures
 A QSAR model developed by Prof. Breneman’s group in Rensselaer
Polytechnic Institute *
 Isozyme specific, 9 available
– Cytochrome P450 isozymes 1A2, 2A6, 2B6, 2C19, 2C8, 2C9, 2D6, 2E1, and 3A4
 Academic scientific project with technical limitations
– Command line only, no graphic interface
– Structure files input/output – bad for chemists
– Single user only, not capable of running predictions simultaneously
– Dependencies on MOE and MatLab
* Jed Zaretzki, Patrik Rydberg, Charles Bergeron, Kristin P. Bennett, Lars Olsen, and Curt M. Breneman, J. Chem. Inf. Model. 2012, 52,
1637−1659

7. RS Predictor at Cubist
 Modify the run scripts to manage
– MOE floating licenses
– Matlab license
– Multiple users, not overwrite results
– Simultaneous runs
– Output result folders
 Allow input structure from a drawing package
– MarvinSketch
 Visualize the prediction results
– MarvinView (customized to display atom label and highlight)
More details on integration later

8. Outline
 Introduction to drug metabolism
 Introduction to Cubist Discovery Informatics group
 The DT workbench
 Case study

9. Discovery Informatics
 Not a software development team
– involved in bringing in new technologies
 Part of Discovery Technologies (DT) - a hybrid group consisting of
– chemo- and bio- informatics
– molecular modeling
– compound management
– High Throughput Screening
– NMR / Biophysics
 Every cheminformatician supports at least one research project
– Driving project SAR
– Generating ideas using computational technologies
– Ensuring the availability & integration of data
– Chemical and biological data management

10. Outline
 Introduction to drug metabolism
 Introduction to Cubist Discovery Informatics group
 The DT workbench
 Case study

11. DT Workbench - background
 Used to publish models for access by chemists
– User friendly, web based instead of script based
– Many expert tools are command line only
– Created, maintained, and expanded by DT scientists, not IT
 Feedback has been very positive
– Started as a prototype in 2008
– Lead to substantial update in 2012
• ChemOffice replaced by ChemAxon Marvin
• Increased usage since updates
 Low maintainence: 3tier web-based system
– Thin client that utilize existing servers and storage disks
– No installation at the user end
 Templates: Consistent but customizable front end
– Apps have different fields depends on inputs parameters
– Templates allow reuse of existing forms
– Common components reduces training requirement
 Scalability: Re-designed system architecture to incorporate
new technologies, such as Marvin suite for both input and
visualization

12. DT workbench – main user interface
Tree menu organizes
tools/models
Consistent layout
Option to retrieve result by
email – for big jobs
• Customizable input form depends
on run parameters
• Templates based system allow
reuse of existing forms
Integrated Marvin Sketch
Applet, no end user
software installation
- especially important for
end users without admin
privileges
ChemAxon components in Pipeline Pilot
used to calculate some properties

13. System Architecture
- the design for DT workbench
13
DT Workbench
MarvinSketch Applet
MarvinView Applet

14. System Architecture
- the design for DT workbench
MarvinSketch Applet
MarvinView Applet
• Communicate back to Tomcat server to send the structure data
• Compatible with Cubist supported browser and platform
• Better fit within W3C standard HTML than ActiveX (ChemDraw)
• Communication between Javascript and Java - enable interactive view

15. Outline
 Introduction to drug metabolism
 Introduction to Cubist Discovery Informatics group
 The DT workbench
 Case study

16. Case study
 Identify potential liability of Ticrynafen by predicting its SOM
 Lead hopping – searching new ideas
 Evaluating the new structure – overlay with Ticrynafen

17. Start with Ticrynafen’s structure
Integrated with Marvin
Sketch for structure
drawing
Cytochrome P450 isozyme
model to use

18. Running prediction on the server
Change your mind
after job submitted?
You can still get the
results by email
Information about the
Job: task name, and
status

19. Predicted results – identified site of metabolism
Customized MarvinView to visualize predicted
results
- Critical information is directly displayed on
the structure
- Thanks to Marvin API, difficult to do this
with ChemDraw
Predicted result is from scripts in sdf file format
– no interface
Thiophene is
the main site of
metabolism
Part of MarvinView,
automatically display relevant
properties based on model
run
Highlight the
primary SOM
Label the SOM
atoms

20. Predicted results – identified site of metabolism
SOM_viewer.jsp
• Simple coding with Marvin API
• Self-taught by reading ChemAxon Documents

21. Case studies
 Identify potential liability of Ticrynafen by predicting its SOM
 Lead hopping – searching new ideas
 Evaluating the new structure – overlay with Ticrynafen

22. Thiophene replacement – Lead hopping
Take the output of current run into the next job
Thiophene was
identified as primary site
of metabolism so we
want to replace it

23. Thiophene replacement – Lead hopping
Find replacement with similar biological activity
Similar Template

24. Thiophene replacement – Lead hopping
2D structure viewer – customized with Marvin view

25. 2D Structure View – Multiple compounds
 Customized interface
 Integrated with Knime
– Knime pre-processes the SDF file
• Standardize structures / tidy file
• Chunk file for multiple page display
 MarvinView API used
– Table layout
– Structure display parameters
– Event listener, etc.
Switch view types Display by pages
Toggle to display/hide
properties
Thiophene
replacements

26. Case studies - continue
 Identify potential liability of Ticrynafen by predicting its SOM
 Lead hopping – searching new ideas
 Evaluating the new structure – overlay with Ticrynafen

27. Evaluate the new structure
slight modification of interface
- same components

28. 3D structure overlay
3D structure viewer – jmol
Needs improvement
MarvinSpace evaluated, but not
adopted due to slow loading of pdb
files in our server environment
2D structure viewer – not relevant here

29. Drug metabolism prediction and drug design
Ticrynafen
(withdrawn in 1980)
Tertiary
C
Primary
O
O
OH
O
S
Secondary
Cl
Cl
Predicted
primary SOM
non-standard bioisostere
Edecrin by Merck
(approved in 1999)
bioisostere

30. Summary
 Metabolism Plays Central Role in Drug Discovery
– Computational models can help
 Integrated Web-based Applications Reduce Learning Curves
– Standard components and templates
 Technologies such as Marvin enable Quick, Inexpensive, Simple Solutions
– 6 month develop by a co/op student from North Eastern university
– Low maintenance - requires no additional IT support
 Marvin Sketch Applet is Consistent but Flexible
– Easily implemented in web pages - no end user installation
– Wide range of file formats allows common interface for different Workbench models
 Marvin View/Structure Renderer is Adaptable to Various Needs
– Highlight specified atoms & bonds
– Add labeling and captioning text
– Interactive Spreadsheet and Grid formats available
 Don’t be Limited by the Out of Box Interfaces
– Marvin customization is easy

31. Acknowledgements
Curt M. Breneman Professor
and Department Head,
Department of Chemistry
and Chemical Biology
Jie Wu
Northeastern Cooperative Student
Dominic Ryan
Senior Director, DT
Cubist Pharmaceuticals
Open-source 3D
structure viewer
THANKS!