Transcript 32) `Advances in data provisioning`

Advances in Data Provisioning
Marian Brodney*, Jacquelyn L. Klug-McLeod, Gregory A. Bakken,
Robert Pfizer World Wide Research and Development
Division of Chemical Information, From Data to Prediction
2016 Spring ACS National Meeting, San Diego, CA
Computational
Sciences CoE
Outline
• Discovery medicinal chemistry is a data driven process
• Discovery project teams use, gather and produce large
amounts of data
• Different kinds of data from different disciplines of the teams
• Strong desire for enabled automated process providing
integrated support for data provisioning, capture and
visualization with user input
• Portfolio solution applied across projects and therapeutic areas
Pfizer Confidential │ 2
Computational
Sciences CoE
Drug Discovery Process
Image: http://www.slideshare.net/PowerPoint-Templates/drug-discovery-process-style-5-powerpoint-presentation-templates
With each phase in the drug discovery process there is data generated. Different
varieties of data, often stored in different data sources feed the progress along the
discovery pathway; ALL DATA generated is used in decision making
Pfizer Confidential │ 3
Computational
Sciences CoE
Drug Discovery Data Sources
Synthesis
information
PubMed
Patents
Competitive
Intetel
Literature
In Vivo Assay
details
In vivo Assay
details
In vivo Assay
data
In Vivo Assay
details
Gene Table
Compiled information
Into user-friendly
format(s)
Targets
BioInfo
NIH
Cortellis
CheMBL
Computed
properties
HTS
Comp
models
Machine
learning
Vendor
Dose
Models
Compound
properties
Design
information
PDM
Models
Compound
tracking
Pfizer Confidential │ 4
Safety
Computational
Sciences CoE
Discovery MedChem Cycle Data Challenges
Design Cycle
• Difficult for discovery project teams to extract
and collate all relevant data of interest
• Data from different sources and in various
formats
• Aggregation rules
• Data formatting/munging
• Query times extend with the number of data
sources used
• Calculations/computational tools used
on full data set, extends query time
• Typical project information file can have
200-500 columns of data for several
thousands of compounds
• Several hours to run
• Teams need relevant and up-to-date data at
their fingertips provided in user-friendly
format for daily use to drive project
progression
Pfizer Confidential │ 5
Computational
Sciences CoE
Solution: Automated Data Presentation (ADP)
ADP : project data driven file providing foundation for all aspects of project informatics
enabling links between design, synthesis and technology and strengthening these
connections through rapid information exchange.
ADP files are the foundation of data compiled to support medchem project teams
• Support solutions to ongoing medicinal chemistry
problems
Design
Analyt.
Chem.
Synthes
is
• Examples: compound summaries, compound and
design tracking, property/data visualizations
• Enable analytics for discovery project teams
Infrastr.
ADP
Struct.
Biology
Cpd.
Safety
Chem.
Biology
Comp.
Sci.
• Examples: R group de-convolutions, series
annotation, pair-wise analysis (MMP)
• Enables solutions to enhance project work flow
• Examples: linking design objectives to virtual
compounds, virtual compounds to synthesized
compounds, design cycles, synthesis queue workflows
• Global Project infrastructure
• Examples: global level alerts and data support to inhouse tools, enables reporting across projects for full
portfolio view
Pfizer Confidential │ 6
Computational
Sciences CoE
Definition of Data Types
• Project specific data
• Any assay data not covered by common data (outlined below)
• In vivo, in vitro, PDM, synthesis tracking, etc.
• Common data
• Assays used across all projects
• Platform data (ADME, CEREP, KSS panel, etc.)
• Project specific calculations
• Series definitions, R-Group deconvolutions, dose models
• Common Calculations
• Used across all projects
• global cADME models, cLogP, cLogD, compound computed
properties, etc.
Pfizer Confidential │ 7
Computational
Sciences CoE
ADP: Definition
• ADP
• Automated Data Presentation
• Retrieval of assay data for real (synthesized) compounds
• Calculations
• Property and other calculations
• Series annotations, RGroup deconvolution, LipE, etc.
• All relevant data pulled from various sources and merged on
compound level updated on a set schedule
• VCP
• Virtual Compound Presentation
• Calculations for virtual compounds
• Functionality for VCP is a subset of what is needed for ADP
Pfizer Confidential │ 8
Computational
Sciences CoE
ADP: Version 1
Internally Developed Solution
• A generalized automated process developed by cheminformatics group using
protocols written in Pipeline Pilot to retrieve all compounds and relevant data of
interest to a team to enable project progression
• Scheduled output results (includes most recent data), refreshed as often as
data updated (every 15 min)
• Addresses problem of speed/stability
• Driven by team-modifiable input (assay dependant, project code dependent)
• Various aggregation levels
• Customizable—logic, sorting, filtering, annotations, external data
sources, etc
• Project specific and global platform data
• Compounds can be synthesized and/or a combination of reals and virtuals
• Data is available in various visualization tools
Pfizer Confidential │ 9
Computational
Sciences CoE
ADP: Version 1
Problems:
Complicated process
managed by 4-6
people
Supporting ~65
teams
Project teams did not
have DIRECT control
of data
Refreshes/changes
upon request
Pfizer Confidential │ 10
Computational
Sciences CoE
ADP: Version 2
Vendor Developed Solution
• Teams have more direct control of their data in D360
• Selection of aggregation category
• Logic/sorting/equations/annotations
• Presentation of data set output in customizable forms or lists (conditional
coloring, sorting, manipulation, etc)
• Can schedule output results to update nightly
•
•
•
•
Driven by team-modifiable input (assay dependant, project code dependent)
Project specific and global platform data
Compounds can be synthesized (reals)
or a combination of reals and virtuals
Problems:
Data is available in various visualization
tools
Lack of
Stability
Inconsistent capability with internal tools/systems
Long running/large queries not well supported
Short scheduling window (1x day)
No direct export of files
Pfizer Confidential │ 11
Teams not able to get the data sets
they need in a recent amountComputational
of time
Sciences CoE
ADP: Version 3
Hybrid Developed Solution
• Using D360 as front end for project-initiated queries
• Project teams still OWN their D360 queries
• Direct control of data input/level of aggregation, etc.
• Logic/sorting/equations/annotations/customizations
• Schedule to run nightly
• Can update/refresh as needed (via schedule window)
• To minimize complexity of the queries, specify only project specific
information in D360 queries
• Queries for ALL desired data not supportable.
• Internal Pfizer team (CSCoE) developed a data provisioning platform to pick
up D360 query files, add additional global data and deposit the full project
files into shared folders for direct application access
• Global/platform data provided via the secondary PLP process
• Full project
files provided
eachProcess
D360 refresh
Combined
benefitsto
ofteams
vendorand
andupdated
Legacy with
internal
Computational
Sciences CoE
ADP: Version 3
Internal tools
Spotfire 6.5
Pfizer Confidential │ 13
Tools reconfigured for uniform access to data
Computational
Sciences CoE
ADP: Version 3
D360 Scheduled Query
1. D360 – nightly scheduled query to bring in
new data, but the same query can be manually
run from the scheduled queries dialog if
desired (immediate access to new data)
Pipeline Pilot
2. Pipeline Pilot – retrieves data file from
D360 system and can add/manipulate
data file as desired, including adding
platform data. Final flat file saved to a
network location.
Taken from: https://community.accelrys.com/thread/4959
Spotfire
3. Upon opening Spotfire DXP (or tool of
choice), the user is presented with
previously saved views (can be incredibly
complex) of the updated data file allowing
for in depth analysis
Computational
Sciences CoE
Taken from: http://www.pressebox.com/attachment/34940/Spotfire_DXP_1.1.png
ADP Visualizations
View of Kinase panel data at various doses
(1, 10, 50uM) on Kinome Tree for specific
project
Detail view
highlights
potency/intrinsic
activity relationship
for selected
headpiece
MPO score
facilitates
pairwise analysis
in understanding
the SAR around
core modification
Computational
Sciences CoE
ADP Visualizations (cont.)
Tracking project
external vendor queue
in relation to project
Multi-priority.
design
parameter
Summary:
Average
property trends
for 2 series
relative to
Tracking project
stated
design
external/internal
objectives
synthesis queue
colored by source.
Computational
Sciences CoE
ADP Visualizations (cont.)
MPO-NSG
facilitates
identification of
SAR clusters with
optimal property
space alignment
(clusters E and
F) and those with
in continuous,
low MPO space
(clusters A, B, C,
and D).
Computational
Sciences CoE
ADP Visualizations (cont.) Pfizer’s BACE series
Effect of log D and pKa on hERG IC50.
(A) Diverse Pfizer set of 2044 compounds. (B) Set of 169 BACE compounds from property space I
and II. Red, hERG IC50 < 10 μM; blue, hERG IC50 > 10 μM. Total count per bin is highlighted in
the center of the pie.
Brodney et al J. Med. Chem. 2015, 58, 3223−3252
Computational
Sciences CoE
ADP Visualizations (cont.) Pfizer’s Early LpxC Series
Goals – remove alkyne, reduce
clearance, improve solubility
Results – attractive series with wild type Pae in vivo
activity, but challenging synthesis and limited
spectrum.
Warmus et al., BOMCL (2012), 22(7), 2536
Computational
Sciences CoE
ADP Visualizations: Portfolio Roll Up
Interactive DXP Dashboard tool
• Project ADP files enable portfolio level tool
• Track project progression over time across zones
• Identify common issues for collaboration
Help to identify
project bottlenecks as well as
highlight
efficiencies
Computational
Sciences CoE
Summary
• Drug discovery is data driven process
• Essential to have all relevant information for decision making
• CSCoE group developed an automated global platform process to
provide all relevant data to project teams as ADP files
• ADP files provide the foundation for all aspects of project
informatics enabling links between design, synthesis and
technology and strengthening these connections through
rapid information exchange
Computational
Sciences CoE
Acknowledgments
Pfizer Global Research and Development: Departments
•
•
•
•
Computational Sciences Center of Emphasis (CSCoE)
Cheminformatics (legacy)
Discovery Medicinal Chemistry (WWMC)
Business Technology (BT)
Pfizer Global Research and Develop Project Teams
IDE Development Team
Global ADP team
External Companies
• Accelrys/Bovia
• Cetera
• Tibco
Pfizer Confidential │ 22
Computational
Sciences CoE
Acknowledgments
•Groton ChemInformatics/DADA
•Artie Brosius
•Marian Brodney
•Chris Poss
•Steve Heck
•Tracy Gregory
•Jacquelyn Klug-McLeod
•Alan Mathiowetz
•Brian Bronk
•Jared Milbank
•Accelrys/Bovia
•Andrei Caracoti
•Dimitri Bondarev
•Klaus Dress
•Bruce Lefker
•Greg Bakken
•Tien Sng
•Brock Luty
•Lourdes Cucurull-Sanchez
•Mike Linhares
•Josh Du
•Veer Shanmugasundaram
•Rob Stanton
•Chris Kibbey
•Steve Rieth
•Justin Montgomery
•Robert Owen
•Bruce Rogers
Pfizer Confidential │ 23
Computational
Sciences CoE
Back Up Slides
Pfizer Confidential │ 24
Computational
Sciences CoE
ADP Visualizations (cont.) Pyridone Methylsulfone Hydroxamates



More polar but still active cores discovered.
General trend of increasing free
fraction with increasing polarity
MICs drop off if cLogD drops
below 0
Free fraction too low if cLogD is
too far above 1
Montgomery et al., J. Med. Chem. 2012, 55, 1662.
Computational
Sciences CoE