Collaborative Database and Computational Models for Tuberculosis Drug Discovery

Collaborative Database and Computational Models for Tuberculosis Drug Discovery Sean Ekins Collaborations in Chemistry, Fuquay Varina, NC. Collaborative Drug Discovery, Burlingame, CA. Department of Pharmacology, University of Medicine & Dentistry of New Jersey-Robert Wood Johnson Medical School, Piscataway, NJ. School of Pharmacy, Department of Pharmaceutical Sciences, University of Maryland, Baltimore, MD.

In the long history of human kind (and animal kind, too) those who have learned to collaborate and improvise most effectively have prevailed. Charles Darwin

Outline
Introduction
Collaborative Drug Discovery
TB Collaborations and Drug Discovery Research
Open ADME Models
Repurposing FDA approved drugs
The Future – Mobile Apps for Drug Discovery

Open Innovation Open innovation is a paradigm that assumes that firms can and should use external ideas as well as internal ideas, and internal and external paths to market, as the firms look to advance their technology Chesbrough, H.W. (2003). Open Innovation: The new imperative for creating and profiting from technology. Boston: Harvard Business School Press, p. xxiv Collaborative Innovation A strategy in which groups partner to create a product - drive the efficient allocation of R&D resources. Collaborating with outsiders-including customers, vendors and even competitors-a company is able to import lower-cost, higher-quality ideas from the best sources in the world. Open Source While open source and open innovation might conflict on patent issues, they are not mutually exclusive, as participating companies can donate their patents to an independent organization, put them in a common pool or grant unlimited license use to anybody. Hence some open source initiatives can merge the two concepts Some Definitions

How to do it better? What can we do with software to facilitate it ? The future is more collaborative We have tools but need integration
Groups involved traverse the spectrum from pharma, academia, not for profit and government
More free, open technologies to enable biomedical research
Precompetitive organizations, consortia..
A starting point for collaboration A core root of the current inefficiencies in drug discovery are due to organizations’ and individual’s barriers to collaborate effectively Bunin & Ekins DDT 16: 643-645, 2011

Major collaborative grants in EU: Framework, IMI …NIH moving in same direction? Cross continent collaboration CROs in China, India etc – Pharma’s in US / Europe More industry – academia collaboration ‘not invented here’ a thing of the past More effort to go after rare and neglected diseases -Globalization and connectivity of scientists will be key – Current pace of change in pharma may not be enough. Need to rethink how we use all technologies & resources… Collaboration is everywhere

Hardware is getting smaller 1930’s 1980s 1990s Room size Desktop size Not to scale and not equivalent computing power – illustrates mobility Laptop Netbook Phone Watch 2000s

Models and software becoming more accessible- free, precompetitive efforts - collaboration Free tools are proliferating

Typical Lab: The Data Explosion Problem & Collaborations DDT Feb 2009

Collaborative Drug Discovery Platform
CDD Vault – Secure web-based place for private data – private by default
CDD Collaborate – Selectively share subsets of data
CDD Public – public data sets - Over 3 Million compounds, with molecular properties, similarity and substructure searching, data plotting etc
will host datasets from companies, foundations etc
vendor libraries (Asinex, TimTec, ChemBridge)
Unique to CDD – simultaneously query your private data, collaborators’ data, & public data, Easy GUI
www.collaborativedrug.com

CDD: Single Click to Key Functionality

CDD: Mining across projects and datasets

Tuberculosis Kills 1.6-1.7m/yr (~1 every 8 seconds)
1/3 rd of worlds population infected!!!!
Multi drug resistance in 4.3% of cases
Extensively drug resistant increasing incidence
No new drugs in over 40 yrs
Drug-drug interactions and Co-morbidity with HIV
Collaboration between groups is rare
These groups may work on existing or new targets
Use of computational methods with TB is rare
Literature TB data is not well collated (SAR)
Funded by Bill and Melinda Gates Foundation
Applying CDD to Build a disease community for TB

~ 20 public datasets for TB Including Novartis data on TB hits >300,000 cpds Patents, Papers Annotated by CDD Open to browse by anyone http://www.collaborativedrug.com/register Molecules with activity against

CDD is a partner on a 5 year project supporting >20 labs and proving cheminformatics support www.mm4tb.org More Medicines for Tuberculosis

Ekins et al, Trends in Microbiology 19: 65-74, 2011 Fitting into the drug discovery process

Searching for TB molecular mimics; collaboration Lamichhane G, et al Mbio, 2: e00301-10, 2011 Modeling – CDD Biology – Johns Hopkins Chemistry – Texas A&M

Simple descriptor analysis on > 300,000 compounds tested vs TB 4.72 (1.99) 77.75 (30.17)** 42.43 (8.94)* 0.12 (0.34)** 4.24 (1.58) 1.11 (0.82)** 3.38 (1.36)** 352.59 (70.87) Inactive < 90% inhibition at 10uM (N =100,931) 4.76 (1.99) 70.28 (29.55) 41.88 (9.44) 0.19 (0.40) 4.18 (1.66) 0.98 (0.84) 4.04 (1.02) 349.58 (63.82) Active ≥ 90% inhibition at 10uM (N =1702) TAACF-NIAID CB2 4.91 (2.35) 85.06 (32.08)* 43.38 (10.73) 0.09 (0.31)** 4.86 (1.77) 1.14 (0.88) 2.82 (1.44)** 350.15 (77.98)** Inactive < 90% inhibition at 10uM (N = 216367) 4.85 (2.43) 83.46 (34.31) 42.99 (12.70) 0.20 (0.48) 4.89 (1.94) 1.16 (0.93) 3.58 (1.39) 357.10 (84.70) Active ≥ 90% inhibition at 10uM (N = 4096) MLSMR RBN PSA Atom count RO 5 HBA HBD logP MWT Dataset

Bayesian Classification Models for TB Good Bad active compounds with MIC < 5uM Laplacian-corrected Bayesian classifier models were generated using FCFP-6 and simple descriptors. 2 models 220,000 and >2000 compounds Ekins et al., Mol BioSyst, 6: 840-851, 2010

Bayesian Classification Dose response Good Bad Ekins et al., Mol BioSyst, 6: 840-851, 2010

Bayesian machine learning Ekins, Williams and Xu, Drug Metab Dispos 38: 2302-2308, 2010 Bayesian classification is a simple probabilistic classification model. It is based on Bayes’ theorem h is the hypothesis or model d is the observed data p ( h ) is the prior belief (probability of hypothesis h before observing any data) p ( d ) is the data evidence (marginal probability of the data) p ( d|h ) is the likelihood (probability of data d if hypothesis h is true) p ( h|d ) is the posterior probability (probability of hypothesis h being true given the observed data d ) A weight is calculated for each feature using a Laplacian-adjusted probability estimate to account for the different sampling frequencies of different features. The weights are summed to provide a probability estimate

Bayesian Classification TB Models Leave out 50% x 100 Ekins et al., Mol BioSyst, 6: 840-851, 2010 65.47 ± 7.96 67.21 ± 7.05 66.85 ± 4.06 0.75 ± 0.01 0.73 ± 0.01 MLSMR dose response set (N = 2273) 77.13 ± 2.26 78.59 ± 1.94 78.56 ± 1.86 0.86 ± 0 0.86 ± 0 MLSMR All single point screen (N = 220463) Sensitivity Specificity Concordance Internal ROC Score External ROC Score Dateset (number of molecules)

100K library Novartis Data FDA drugs Additional test sets Suggests models can predict data from the same and independent labs Initial enrichment – enables screening few compounds to find actives 21 hits in 2108 cpds 34 hits in 248 cpds 1702 hits in >100K cpds Ekins and Freundlich, Pharm Res, 28, 1859-1869, 2011. Ekins et al., Mol BioSyst, 6: 840-851, 2010

Bayesian Models

Generated with kinase data [1] - - (blind testing of previous models showed 3-4 fold
enrichment )

Models were built as described previously [2]

Data for single point screening (cut off for activity % inhibition at 10uM >or equal to
90%)

2.IC 50 data Cut off for active = or equal to 5uM

3.IC 90 data Cut off for active = or equal to 10uM and vero cell selectivity index greater or
equal to 10.

[1] Reynolds RC, et al. Tuberculosis (Edinburgh, Scotland) 2011 In Press.
[2] Ekins S, et al.,Mol BioSystems 2010;6:840-51.

Models with SRI kinase library data

Models with SRI kinase library data; refining data with cytotoxicity Model 1 ROC XV AUC (N 23797) = 0.89 Model 2 (N 1248) = 0.72 Model 3 (N 1248) = 0.77 Leave out 50% x 100 Adding cytotoxicity data improves models Dateset (number of molecules) External ROC Score Internal ROC Score Concordance Specificity Sensitivity Model 1 (N = 23797) 0.87 ± 0 0.88 ± 0 76.77 ± 2.14 76.49 ± 2.41 81.7 ± 2.96 Model 2 (N = 1248) 0.65 ± 0.01 0.70 ± 0.01 61.58 ± 1.56 61.85 ± 8.45 61.30 ± 8.24 Model 3 (N=1248) 0.74 ± 0.02 0.75 ± 0.02 68.67 ± 6.88 69.28 ± 9.84 64.84 ± 12.11

Original TB Models : refining data with cytotoxicity Ekins et al., Mol BioSyst, 6: 840-851, 2010 Single pt ROC XV AUC = 0.88 Dose resp = 0.78 Dose resp + cyto = 0.86 Leave out 50% x 100 Dateset (number of molecules) External ROC Score Internal ROC Score Concordance Specificity Sensitivity MLSMR All single point screen (N = 220463) 0.86 ± 0 0.86 ± 0 78.56 ± 1.86 78.59 ± 1.94 77.13 ± 2.26 MLSMR dose response set (N = 2273) 0.73 ± 0.01 0.75 ± 0.01 66.85 ± 4.06 67.21 ± 7.05 65.47 ± 7.96 NEW Dose resp and cytotoxicity (N = 2273) 0.82 ± 0.02 0.84 ± 0.02 82.61 ± 4.68 83.91 ± 5.48 65.99 ± 7.47

Combining cheminformatics methods and pathway analysis
Identified essential TB targets that had not been exploited
Used resources available to both to identify targets and molecules that mimic substrates
Computationally searched >80,000 molecules - tested 23 compounds in vitro (3 picked as inactives), lead to 2 proposed as mimics of D-fructose 1,6 bisphosphate, (MIC of 20 and 40 ug/ml)
POC took < 6mths - - Submitted phase II STTR, Submitted manuscript
Still need to test vs target - verify it hits suggested target
Ekins et al, Trends in Microbiology Feb 2011 Phase I STTR - NIAID funded collaboration with Stanford Research International

http://www.slideshare.net/ekinssean Ekins S and Williams AJ, MedChemComm, 1: 325-330, 2010. Analysis of malaria and TB data

TB Compound libraries and filter failures Filtering using SMARTs filters to remove thiol reactives, false positives etc at University of New Mexico (http://pasilla.health.unm.edu/tomcat/biocomp/smartsfilter) Ekins et al., Mol Biosyst, 6: 2316-2324, 2010

Antimalarial Compound libraries and filter failures Ekins and Williams Drug Disc Today 15; 812-815, 2010 % Failure

Correlation between the number of SMARTS filter failures and the number of Lipinski violations for different types of rules sets with FDA drug set from CDD (N = 2804) Suggests # of Lipinski violations may also be an indicator of undesirable chemical features that result in reactivity Filter Correlations with Rule of 5 Ekins and Freundlich, Pharm Res, 28, 1859-1869, 2011.

Summary Computational models based on Whole cell TB data could improve efficiency of screening Collaborations get us to interesting compounds quickly Availability of datasets enable analysis that could suggest simple rules A high proportion of compounds fail the Abbott filters for reactivity when compared to drugs and antimalarials Understanding the chemical properties and characteristics of compounds = better compounds for lead optimization.

Could all pharmas share their data as models with each other? Increasing Data & Model Access Ekins and Williams, Lab On A Chip, 10: 13-22, 2010.

Bridging the Gap
Challenge..There is limited access to ADME/Tox data and models needed for R&D
How could a company share data but keep the structures proprietary?
Sharing models means both parties use costly software
What about open source tools?
Pfizr had never considered this - So we proposed a study and Rishi Gupta generated models

What can be developed with very large training and test sets?
HLM training 50,000 testing 25,000 molecules
training 194,000 and testing 39,000
MDCK training 25,000 testing 25,000
MDR training 25,000 testing 18,400
Open molecular descriptors / models vs commercial descriptors
Gupta RR, et al., Drug Metab Dispos, 38: 2083-2090, 2010 Open source tools for modeling

Massive Human liver microsomal stability model PCA of training (red) and test (blue) compounds Overlap in Chemistry space Gupta RR, et al., Drug Metab Dispos, 38: 2083-2090, 2010
# Descriptors: 818 Descriptors
# Training Set compounds: 193,930
Cross Validation Results: 38,786 compounds
Training R 2 : 0.77
20% Test Set R 2 : 0.69
Blind Data Set (2310 compounds):
R 2 = 0.53
RMSE = 0.367
Continuous  Categorical:
κ = 0.42
Sensitivity = 0.24
Specificity = 0.987
PPV = 0.823
Time (sec/compound): 0.303
# Descriptors: 578 Descriptors
# Training Set compounds: 193,650
Cross Validation Results: 38,730 compounds
Training R 2 : 0.79
20% Test Set R 2 : 0.69
Blind Data Set (2310 compounds):
R 2 = 0.53
RMSE = 0.367
Continuous  Categorical:
κ = 0.40
Sensitivity = 0.16
Specificity = 0.99
PPV = 0.80
Time (sec/compound): 0.252
HLM Model with MOE2D and SMARTS Keys HLM Model with CDK and SMARTS Keys:

RRCK Permeability and MDR Open descriptors results almost identical to commercial descriptors Across many datasets and quantitative and qualitative data Smaller solubility datasets give similar results Provides confidence that open models could be viable MDCK training 25,000 testing 25,000 MDR training 25,000 testing 18,400 Gupta RR, et al., Drug Metab Dispos, 38: 2083-2090, 2010 Kappa = 0.50 Sensitivity = 0.62 Specificity = 0.94 PPV = 0.68 Kappa = 0.53 Sensitivity = 0.64 Specificity = 0.94 PPV = 0.72 (Baseline) Kappa = 0.47 Sensitivity = 0.59 Specificity = 0.93 PPV = 0.67 C5.0 RRCK Permeability Kappa = 0.65 Sensitivity = 0.86 Specificity = 0.78 PPV = 0.84 CDK and SMARTS Keys Kappa = 0.67 Sensitivity = 0.86 Specificity = 0.80 PPV = 0.85 (Baseline) MOE2D and SMARTS Keys Kappa = 0.62 Sensitivity = 0.85 Specificity = 0.77 PPV = 0.83 CDK descriptors C5.0 MDR

Merck KGaA Combining models may give greater coverage of ADME/ Tox chemistry space and improve predictions? Model coverage of chemistry space Lundbeck Pfizer Merck GSK Novartis Lilly BMS Allergan Bayer AZ Roche BI Merk KGaA

Next steps
ADME/Tox Data crosses diseases
Potential to share models selectively with collaborators e.g. academics, neglected disease researchers
We used the proof of concept to submit an SBIR “ Biocomputation across distributed private datasets to enhance drug discovery”
Develop prototype for sharing models securely- collaborate to show how combining data for TB etc could improve models
Phase II- develop a commercial product that leverages CDD

Bunin & Ekins DDT 16: 643-645, 2011 A complex ecosystem of collaborations: A new business model Inside Company Collaborators Inside Academia Collaborators Molecules, Models, Data Molecules, Models, Data Inside Foundation Collaborators Molecules, Models, Data Inside Government Collaborators Molecules, Models, Data IP IP IP IP Shared IP Collaborative platform/s

Finding Promiscuous Old Drugs for New Uses
Research published in the last six years - 34 studies - Screened libraries of FDA approved drugs against various whole cell or target assays.
1 or more compounds with a suggested new bioactivity
13 drugs were active against more than one additional disease in vitro

Finding Promiscuous Old Drugs for New Uses
109 molecules were identified by screening in vitro
Statistically more hydrophobic (log P) and higher MWT than orphan-designated products with at least one marketing approval for a common disease indication or one marketing approval for a rare disease from the FDA’s rare disease research database.
Created structure searchable databases in CDD

2D Similarity search with “hit” from screening Export database and use for 3D searching with a pharmacophore or other model Suggest approved drugs for testing - may also indicate other uses if it is present in more than one database Suggest in silico hits for in vitro screening Key databases of structures and bioactivity data FDA drugs database Repurpose FDA drugs in silico Ekins S, Williams AJ, Krasowski MD and Freundlich JS, Drug Disc Today, 16: 298-310, 2011

Crowdsourcing Project “Off the Shelf R&D” All pharmas have assets on shelf that reached clinic “ Off the Shelf R&D” Get the crowd to help in repurposing / repositioning these assets How can software help? - Create communities to test - Provide informatics tools that are accessible to the crowd - enlarge user base - Data storage on cloud – integration with public data - Crowd becomes virtual pharma-CROs and the “customer” for enabling services

Tools for Open Science
Blogs
Wikis
Databases
Journals
What about Twitter, Facebook, could these be used for social collaboration, science?

2020: A Drug Discovery Odyssey Could our Pharma R&D look like this Massive collaboration networks – software enabled. We are in “Generation App”. Crowdsourcing will have a role in R&D. Drug discovery possible by anyone with “app access” Ekins & Williams, Pharm Res, 27: 393-395, 2010.

Make science more accessible = >communication
Mobile – take a phone into field /lab and do science more readily than on a laptop
GREEN – energy efficient computing
MolSync + DropBox + MMDS = Share molecules as SDF files on the cloud = collaborate
Mobile Apps for Drug Discovery Williams et al DDT 16:928-939, 2011

www.scimobileapps.com How do you find scientific mobile Apps ? Development of Wiki’s to track developments in tools..

Acknowledgments
Antony J. Williams (RSC)
Rishi Gupta , Eric Gifford, Ted Liston, Chris Waller (Pfizer)
Joel Freundlich (Texas A&M), Gyanu Lamichhane (Johns Hopkins)
Carolyn Talcott, Malabika Sarker , Peter Madrid, Sidharth Chopra (SRI International)
MM4TB colleagues
Chris Lipinski
Takushi Kaneko (TB Alliance)
Nicko Goncharoff (SureChem)
Matthew D. Krasowski (University of Iowa)
Alex Clark (Molecular Materials Informatics, Inc)
Accelrys
CDD – Barry Bunin
Funding BMGF, NIAID.
Email: ekinssean@yahoo.com
Slideshare: http://www.slideshare.net/ekinssean
Twitter: collabchem
Blog: http://www.collabchem.com/
Website: http://www.collaborations.com/CHEMISTRY.HTM

Novartis aerobic and anaerobic TB hits Anaerobic compounds showed statistically different and higher mean descriptor property values compared with the aerobic hits (e.g. molecular weight, logP, hydrogen bond donor, hydrogen bond acceptor, polar surface area and rotatable bond number) The mean molecular properties for the Novartis compounds are in a similar range to the MLSMR and TAACF-NIAID CB2 hits Ekins and Freundlich, Pharm Res, 28, 1859-1869, 2011.

Collaborative Database and Computational Models for Tuberculosis Drug Discovery

by Sean Ekins

Accessibility

Categories

Upload Details

Usage Rights

2 Embeds 2

Statistics

Collaborative Database and Computational Models for Tuberculosis Drug Discovery Presentation Transcript

http://a0.twimg.com	1
http://paper.li	1