SOT short course on computational toxicology

Sean Ekins, M.Sc, Ph.D., D.Sc.

Collaborations in Chemistry, Fuquay-Varina, NC.

Collaborative Drug Discovery, Burlingame, CA.School of Pharmacy, Department of Pharmaceutical

Sciences, University of Maryland. 215-687-1320

[email protected]

Computational Models for

Predicting Human Toxicities

• Key enablers

• What has been modeled – a quick review

• What will be modeled

• Future

Outline

Why Use Computational Models For Toxicology?

Goal of a model – Alert you to potential toxicity, enable you to focus efforts on best molecules – reduce risk

Selection of model – trade off between interpretability, insights for modifying molecules, speed of calculation and coverage of chemistry space – applicability domain

Models can be built with proprietary, open and commercial tools

software (descriptors + algorithms) + data = model/s

Human operator decides whether a model is acceptable

Key enablers: 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

Key Enablers: More data available and open tools

• Details

• Details

What has been modeled

• Physicochemical properties, LogP, logD, Solubility, boiling point, melting point

• QSAR for various proteins, complex properties• Homology models, Docking• Expert systems• Hybrid methods – combine different approaches• Mutagenicity (Ames, micronucleus, clastogenicity,

and DNA damage, developmental tox.. )• Environmental Tox – Aquatic, dermatotoxicology• Mixtures – using PBPK

Physicochemical properties• Solubility data – 1000’s data in Literature • Models median error ~0.5 log = experimental error• LogP –tens of 1000’s data available• Fragmental or whole molecule predictors• All logP predictors are not equal. Median error ~ 0.3 log = experimental

error• People now accept solubility and LogP predictions as if real

ACD predictions + EpiSuite predictions in www.chemspider.com

• Mobile molecular data sheet

• Links to melting point predictor from open notebook science

• Required curation of data

Simple Rules• Rule of 5

• Lipinski, Lombardo, Dominy, Feeney Adv. Drug Deliv. Rev. 23: 3-25 (1997).

• AlogP98 vs PSA• Egan, Merz, Baldwin, J. Med. Chem. 43: 3867-3877 (2000)

• Greater than ten rotatable bonds correlates with decreased rat oral bioavailability• Veber, Johnson, Cheng, Smith, Ward, Kopple. J Med Chem 45: 2515–2623, (2002)

• Compounds with ClogP < 3 and total polar surface area > 75A2 fewer animal toxicity findings.

• Hughes, et al. Bioorg Med Chem Lett 18, 4872-4875 (2008).

L. Carlsson,et al., BMC Bioinformatics 2010, 11:362

MetaPrint 2D in Bioclipse- free metabolism site predictor

Uses fingerprint descriptors and metabolite database to learn frequencies of metabolites in various substructures

QSAR for Various Proteins

• Enzymes – predominantly Cytochrome P450s - for drug-drug interactions

• Transporters – predominantly P-gp but some others e.g. OATP, BCRP -

• Receptors – PXR, CAR, for hepatotoxicity

• Ion Channels – predominantly hERG for cardiotoxicity

• Issues – initially small training sets – public data is a fraction of what drug companies have

Pharmacophores

Ideal when we have few molecules for training In silico database searching

Accelrys Catalyst in Discovery Studio

Geometric arrangement of functional groups necessary for a biological response

•Generate 3D conformations•Align molecules•Select features contributing to activity•Regress hypothesis•Evaluate with new molecules

•Excluded volumes – relate to inactive molecules

CYP2B6CYP2C9CYP2D6CYP3A4CYP3A5CYP3A7hERGP-gpOATPsOCT1OCT2BCRPhOCTN2ASBThPEPT1hPEPT2FXR LXRCARPXR etc

hOCTN2 – Organic Cation transporterPharmacophore

• High affinity cation/carnitine transporter - expressed in kidney, skeletal muscle, heart, placenta and small intestine

• Inhibition correlation with muscle weakness - rhabdomyolysis• A common features pharmacophore developed with 7 inhibitors• Searched a database of over 600 FDA approved drugs - selected drugs for in vitro testing. • 33 tested drugs predicted to map to the pharmacophore, 27 inhibited hOCTN2 in vitro

• Compounds were more likely to cause rhabdomyolysis if the Cmax/Ki ratio was higher than 0.0025

Diao, Ekins, and Polli, Pharm Res, 26, 1890, (2009)

hOCTN2 – Organic Cation transporterPharmacophore

Diao, Ekins, and Polli, Pharm Res, 26, 1890, (2009)

• QSAR Examples

Gupta RR, et al., Drug Metab Dispos, 38: 2083-2090, 2010

• Examples – P-gp

Gupta RR, et al., Drug Metab Dispos, 38: 2083-2090, 2010

Open source descriptors CDK and C5.0 algorithm

~60,000 molecules with P-gp efflux data from Pfizer

MDR <2.5 (low risk) (N = 14,175) MDR > 2.5 (high risk) (N = 10,820)

Test set MDR <2.5 (N = 10,441) > 2.5 (N = 7972)

Could facilitate model sharing?

CDK +fragment descriptors MOE 2D +fragment descriptorsKappa 0.65 0.67

sensitivity 0.86 0.86specificity 0.78 0.8

PPV 0.84 0.84

Time dependent inhibition for P450 3A4

• Pfizer generated a large dataset (~2000 compounds) and went through sequential Bayesian model generation and testing cycles

Test set 2 20 active in 156 compounds Combined both model predictions

Zientek et al., Chem Res Toxicol 23: 664-676 (2010)

• 3A4 TDI

Indazole ring, the pyrazole, and the methoxy-aminopyridine rings areimportant for TDI

Approach decreased in vitro screening 30%

Helps identify reactive metabolite forming compounds

Zientek et al., Chem Res Toxicol 23: 664-676 (2010)

• Drug Induced Liver Injury Models

• 74 compounds - classification models (linear discriminant analysis, artificial neural networks, and machine learning algorithms (OneR)) – Internal cross-validation (accuracy 84%, sensitivity 78%, and specificity 90%). Testing

on 6 and 13 compounds, respectively > 80% accuracy.

(Cruz-Monteagudo et al., J Comput Chem 29: 533-549, 2008).

• A second study used binary QSAR (248 active and 283 inactive) Support vector machine models – – external 5-fold cross-validation procedures and 78% accuracy for a set of 18

compounds

(Fourches et al., Chem Res Toxicol 23: 171-183, 2010).

• A third study created a knowledge base with structural alerts from 1266 chemicals. – Alerts created were used to predict results for 626 Pfizer compounds (sensitivity of

46%, specificity of 73%, and concordance of 56% for the latest version) (Greene et al., Chem Res Toxicol 23: 1215-1222, 2010).

• DILI Model - Bayesian

• Laplacian-corrected Bayesian classifier models were generated using Discovery Studio (version 2.5.5; Accelrys).

• Training set = 295, test set = 237 compounds

• Uses two-dimensional descriptors to distinguish between compounds that are DILI-positive and those that are DILI-negative

– ALogP– ECFC_6 – Apol – logD – molecular weight – number of aromatic rings – number of hydrogen bond acceptors – number of hydrogen bond donors – number of rings – number of rotatable bonds – molecular polar surface area – molecular surface area – Wiener and Zagreb indices

Ekins, Williams and Xu, Drug Metab Dispos 38: 2302-2308, 2010

Extended connectivity fingerprints

• DILI Bayesian

Features in DILI -Features in DILI +

Avoid===Long aliphatic chains, Phenols, Ketones, Diols, -methyl styrene, Conjugated structures, Cyclohexenones, Amides

Test set analysis

Ekins, Williams and Xu, Drug Metab Dispos 38: 2302-2308, 2010

• compounds of most interest – well known hepatotoxic drugs (U.S. Food and Drug Administration

Guidance for Industry “Drug-Induced Liver Injury: Premarketing Clinical Evaluation,” 2009), plus their less hepatotoxic comparators, if clinically available.

What will be modeled

• Mitochondrial toxicity, hepatotoxicity, • More Transporters – MATE, OATPs, BSEP..bigger datasets – driven by

academia• Screening centers – more data – more models • Understanding differences between ligands for Nuclear Receptors

– CAR vs PXR

• Models will become replacements for data as datasets expand (e.g. like logP)

• Toxicity Models used for Green Chemistry

Chem Rev. 2010 Oct 13;110(10):5845-82

….Near FutureWider use of models

New methods

Free tools – need good validation studies

Free databases – need to ensure structures / data are correct (DDT editorial Sept 2011)

Concepts perfected on desktop may migrate to apps e.g. collaboration (MolSync+DropBox) Selective sharing of models

Computational ADME/Tox mobile apps?

More efficient tools

Williams et al DDT in press 2011 Bunin & Ekins DDT 16: 643-645, 2011

Acknowledgments• University of Maryland

– Lei Diao– James E. Polli

• Pfizer– Rishi Gupta– Eric Gifford– Ted Liston– Chris Waller

• Merck– Jim Xu

• Antony J. Williams (RSC)

• Accelrys• CDD

• Email: [email protected]

Slideshare: http://www.slideshare.net/ekinssean

Twitter: collabchem

Blog: http://www.collabchem.com/

Website: http://www.collaborations.com/CHEMISTRY.HTM