Presentation at the Chemical Computing Group UGM in 2011. I describe my use of the MOE/web SOAP server and the model of delivering physico-chemical properties onto the medicinal chemists desktop.
CHUYÊN ĐỀ DẠY THÊM TIẾNG ANH LỚP 11 - GLOBAL SUCCESS - NĂM HỌC 2023-2024 - HK...
BIpredict: MOE/web Server Enabled Delivery of In Silico Properties and Models
1. BIpredict: MOE/web S
BI di MOE/ b Server Enabled D li
E bl d Delivery
of In Silico Properties and Models
David C. Thompson, Ph.D
2. 125 Years of Innovating for Patients and Their Families
Founded 1885
in Ingelheim, Germany
Family-owned Privately-held
g
global company
p y for 125 years
y
Products marketed in $17.7 billion 41,500
150+ countries 2009 net sales employees worldwide
Focus on
142 affiliates
ffili t Human
in
Pharmaceuticals
50 countries
& Animal Health
2
3. One Pill Makes You Larger [*]
My favourite papers from each period:
[1] J. Chem. Phys. 122, 124107 (2005)
[2] J. Chem. Phys. 128, 224103 (2008)
[3] J. Chem. Inf. Model. 49, 1889 (2009)
[4] J. Chem. Inf. Model. 51, 93 (2011)
[*] This slide title brought to you courtesy of Lewis Carroll
4. The Magic of Clip Art
What are we trying to do in the pharmaceutical industry?
[*]
+ =
What are we trying to do as computational scientists working in the pharmaceutical
industry?
+ =
Chemical space
We build models to try and expedite the drug
discovery process
[*] Side effects may include changing hair colour 4
5. Shameless slide reuse … [5]
“All models are wrong, but
some models are useful”
– G. E. P. Box
“…the validity of any given model is of limited
scope, as is the case with any mental construct
that we have about what our molecules are doing,
whether we used a software package or waved our
hands around in the air.” – D. Lowe
Simulation and its discontents, Sherry Turkle, Cambridge, MA: MIT Press (2009)
[5] D. C. Thompson et al. Schrödinger Regional User Meeting, New York, NY 2009
6. Taxonomy of Risk [6]
Risk Uncertainty
Randomness amenable to formal Randomness not amenable to formal
statistical analysis statistical analysis
“ … a given phenomenon may contain several levels of uncertainty at once, with some components being completely
certain and others irreducibly uncertain”
“In fact, we propose that the failure of quantitative models [in economics and finance] is almost always attributable
to a mismatch between the level of uncertainty and the methods used to model it.”
it.
[6] “WARNING: Physics Envy May be Hazardous To Your Wealth!”, A. W. Lo, and M. T. Mueller arXiv:1003.2688v3 [q-fin.RM] 6
7. Okay, now what?
• Focus on physicochemical properties [7, 8]
• Enable scientists through light-weight clients
light weight
• Provide core scientific functionality through web services architecture
+ =
Chemical space
We build models to try and expedite the drug
discovery process
[7] Nature Rev. Drug. Discov. 10, 197 (2011)
[8] Med. Chem. Comm., 2011 (DOI: 10.1039/c1md00017a) 7
10. If it’s going to stay, we might as well use it
A Web Service is a method of communication between two electronic devices over a
network[9]
Example*:
Web Service
[9] http://en.wikipedia.org/wiki/Web_service
* Probably 10
11. BIpredict: An in silico molecular property prediction
framework
Initial requirement: Build a real-time physchem. property calculator engine that could be
used to address project concerns and issues at the medicinal chemistry desktop
Offer multiple interfaces to complement users preferred workflow
Buy or Build?
Proposal: Rapid development of an in-house solution to allow us to focus on optimizing the
interaction between a web services layer and other BI systems and, most importantly, the
y y , p y,
scientists
• Leverage Molecular Operating Environment (MOE), and newly developed
MOE/web SOAP application server technology
pp gy
— Java-based web server
— No Apache setup
M OE M OE
Pipeline
Pilot®
ba tc h SOA P
• Focus on flexibility, ease of deployment, and extensibility
11
12. BIpredict architecture:
How you consume, depends on what you see
MOE BIDATA BIModel Pipelining tools Web Apps. Command Line
(G) (G) (G + A) (G + A) (G) (G +A)
Multiple front-ends
(data
(d t consumers)
)
Single back-end
(data producer)
BIpredict
(web services layer)
Synchronous
General (G):
( ) Advanced (A):
( )
Intended for Abstract
general descriptors for
consumption comp. chem.
Development:
Oct. 2009 – Jan. 2010
Production:
Jan. 2010
Interface determines which descriptors are exposed
12
13. “One of the things about a real-time system is everything
has to be timed out” [10]
single back-end
(data producer) BIpredict
(web services layer)
y
Asynchronous / Batched
Descriptor
classes
l
Scatter
User ID
Job ID
General:
42 descriptors
8 engines
Advanced: Y
3431 descriptors N? Collect output
15 engines Descriptor
names
Development: Gather
Jan.
Jan 2010 – June 2010 Job ID
Production:
June 2010 Packaged output
[10] Bernie Cosell, “Czar of the PDP-1 timesharing system” 13
14. What does this magic look like?
With BIpredict panel
open, workspace is
‘live’
Physicochemical
properties are updated
as molecule is built
l l i b il
Atomistic descriptor
values are appended
directly to the molecule
14
16. Reimplementation of Pfizer CNS Multi-Parameter
Optimization design tool [11,12]
• Driven by the scientists, turnaround of days
• G h 119 literature compounds, visually inspect and triage
Gather 9 li d i ll i d i
• Identify physicochemical property Descriptors
yp y p p y p
• Sybyl clogP 7
• ACD logD (@ pH 7.4) 6
• MOE MW 5
R² = 0.9834
-implementation
• CADD TPSA 4
• MOE Lipinski HB donors 3
• ACD M B i pKa
Most Basic K
Re-
2
1
• Enable model through BIpredict
g p 0
0 1 2 3 4 5 6 7
Literature
[11] ACS Chem. Neurosci., 1, 435 (2010)
[12] Bioorg. Med. Chem. Lett, 18, 4872 (2008) 16
17. Begin at the beginning and go on till you come to the end:
then stop [*]
• All models are wrong
• Expose those models that we think will expedite the
p p
drug discovery process
• Focus on extensible, light-weight, service delivery
, g g , y
[*] This slide title brought to you courtesy of Lewis Carroll