Table of Contents

Cover

Title Page

Copyright
Dedication

List of Contributors

Preface

Acknowledgments

Part I: Computational Methods

Chapter 1: Accessible Machine Learning Approaches for Toxicology
1. 1.1 Introduction
2. 1.2 Bayesian Models
3. 1.3 Deep Learning Models
4. 1.4 Comparison of Different Machine Learning Methods
5. 1.5 Future Work
6. Acknowledgments
7. References
Chapter 2: Quantum Mechanics Approaches in Computational Toxicology
1. 2.1 Translating Computational Chemistry to Predictive Toxicology
2. 2.2 Levels of Theory in Quantum Mechanical Calculations
3. 2.3 Representing Molecular Orbitals
4. 2.4 Hybrid Quantum and Molecular Mechanical Calculations
5. 2.5 Representing System Dynamics
6. 2.6 Developing QM Descriptors
7. 2.7 Rational Design of Safer Chemicals
8. References

Part II: Applying Computers to Toxicology Assessment: Pharmaceutical, Industrial and Clinical

Chapter 3: Computational Approaches for Predicting hERG Activity
1. 3.1 Introduction
2. 3.2 Computational Approaches
3. 3.3 Ligand-Based Approaches
4. 3.4 Structure-Based Approaches
5. 3.5 Applications to Predict hERG Blockage
6. 3.6 Other Computational Approaches Related to hERG Liability
7. 3.7 Final Remarks
8. References
Chapter 4: Computational Toxicology for Traditional Chinese Medicine
1. 4.1 Background, Current Status, and Challenges
2. 4.2 Case Study: Large-Scale Prediction on Involvement of Organic Anion Transporter 1 in Traditional Chinese Medicine-Drug Interactions
3. 4.3 Conclusion
4. Acknowledgment
5. References
Chapter 5: Pharmacophore Models for Toxicology Prediction
1. 5.1 Introduction
2. 5.2 Antitarget Screening
3. 5.3 Prediction of Liver Toxicity
4. 5.4 Prediction of Cardiovascular Toxicity
5. 5.5 Prediction of Central Nervous System (CNS) Toxicity
6. 5.6 Prediction of Endocrine Disruption
7. 5.7 Prediction of ADME
8. 5.8 General Remarks on the Limits and Future Perspectives for Employing Pharmacophore Models in Toxicological Studies
9. References
Chapter 6: Transporters in Hepatotoxicity
1. 6.1 Introduction
2. 6.2 Basolateral Transporters
3. 6.3 Canalicular Transporters
4. 6.4 Data Sources for Transporters in Hepatotoxicity
5. 6.5 In Silico Transporters Models
6. 6.6 Ligand-Based Approaches
7. 6.7 OATP1B1 and OATP1B3
8. 6.8 NTCP
9. 6.9 OCT1
10. 6.10 OCT2
11. 6.11 MRP1, MRP3, and MRP4
12. 6.12 BSEP
13. 6.13 MRP2
14. 6.14 MDR1/P-gp
15. 6.15 MDR3
16. 6.16 BCRP
17. 6.17 MATE1
18. 6.18 ASBT
19. 6.19 Structure-Based Approaches
20. 6.20 Complex Models Incorporating Transporter Information
21. 6.21 In Vitro Models
22. 6.22 Multiscale Models
23. 6.23 Outlook
24. Acknowledgments
25. References
Chapter 7: Cheminformatics in a Clinical Setting
1. 7.1 Introduction
2. 7.2 Similarity Analysis Applied to Drug of Abuse/Toxicology Immunoassays
3. 7.3 Similarity Analysis Applied to Therapeutic Drug Monitoring Immunoassays
4. 7.4 Similarity Analysis Applied to Steroid Hormone Immunoassays
5. 7.5 Cheminformatics Applied to “Designer Drugs”
6. 7.6 Relevance to Antibody-Ligand Interactions
7. 7.7 Conclusions and Future Directions
8. Acknowledgment
9. References

Part III: Applying Computers to Toxicology Assessment: Environmental and Regulatory Perspectives

Chapter 8: Computational Tools for ADMET Profiling
1. 8.1 Introduction
2. 8.2 Cheminformatics Approaches for ADMET Profiling
3. 8.3 Unsolved Challenges in Structure Based Profiling
4. 8.4 Perspectives
5. 8.5 Conclusions
6. Acknowledgments
7. Disclaimer
8. References
Chapter 9: Computational Toxicology and Reach
1. 9.1 A Theoretical and Historical Introduction to the Evolution Toward Predictive Models
2. 9.2 Reach and the Other Legislations
3. 9.3 Annex XI of Reach for QSAR Models
4. 9.4 The ECHA Guidelines and the Use of QSAR Models within ECHA
5. 9.5 Conclusions
6. References
Chapter 10: Computational Approaches to Predicting Dermal Absorption of Complex Topical Mixtures
1. 10.1 Introduction
2. 10.2 Principles of Dermal Absorption
3. 10.3 Dermal Mixtures
4. 10.4 Model Systems
5. 10.5 Local Skin Versus Systemic Endpoints
6. 10.6 QSAR Approaches to Model Dermal Absorption
7. 10.7 Pharmacokinetic Models
8. 10.8 Conclusions
9. References

Part IV: New Technologies for Toxicology, Future Perspectives

Chapter 11: Big Data in Computational Toxicology: Challenges and Opportunities
1. 11.1 Big Data Scenario of Computational Toxicology
2. 11.2 Fast-Growing Chemical Toxicity Data
3. 11.3 The Use of Big Data Approaches in Modern Computational Toxicology
4. 11.4 Challenges of Big Data Research in Computational Toxicology and Relevant Forecasts
5. References
Chapter 12: HLA-Mediated Adverse Drug Reactions: Challenges and Opportunities for Predictive Molecular Modeling
1. 12.1 Introduction
2. 12.2 Human Leukocyte Antigens
3. 12.3 Structure-Based Molecular Docking to Study HLA-Mediated ADRs
4. 12.4 Perspectives
5. References
Chapter 13: Open Science Data Repository for Toxicology
1. 13.1 Introduction
2. 13.2 Open Science Data Repository
3. 13.3 Benefits of OSDR
4. 13.4 Technical Details
5. 13.5 Future Work
6. References
Chapter 14: Developing Next Generation Tools for Computational Toxicology
1. 14.1 Introduction
2. 14.2 Developing Apps for Chemistry
3. 14.3 Green Chemistry
4. 14.4 Polypharma and Assay Central
5. 14.5 Conclusion
6. Acknowledgments
7. References
Index

End User License Agreement

List of Illustrations

Chapter 1: Accessible Machine Learning Approaches for Toxicology

Figure 1.1 Summary of machine learning models generated for Mycobacterium tuberculosis in vitro data. This approach has also been applied to ADME/Tox datasets.
Figure 1.2 Example of Bayesian models implemented in MMDS.
Figure 1.3 (a) A two-layer neural network (one hidden layer of four neurons (or units) and one output layer with two neurons), and three inputs. (b) A three-layer neural network with three inputs, two hidden layers of four neurons each and one output layer. In both cases, there are connections (synapses) between neurons across layers, but not within a layer. Source: Adapted from http://cs231n.github.io/neural-networks-1/.
Figure 1.4 Typical frequency of fingerprints occurrence in the 1024-bin compounds in a dataset.

Chapter 2: Quantum Mechanics Approaches in Computational Toxicology

Figure 2.1 Total mean absolute errors (MAEs) recorded for a selection of popular hybrid density functionals (B3LYP–M06HF), double hybrid functionals (B2PLYP–DSD-BLYP) and two ab initio methods (HF and MP2), reflecting basic physicochemical properties, reaction energetics, and noncovalent interactions from the GMTKN30 database. Results were adopted from a study by Goerigk and Grimme [3].
Figure 2.2 Mean absolute errors (MAEs) recorded for a selection of five semiempirical methods (AM1, PM6, and OM1-3) and two DFT methods (B3LYP and PBE) from the reduced (HCNO elements only) GMTKN24 database [4]. Performance across relevant subsets of the GMTKN24 database is provided in pattern fill next to the total MAEs.
Figure 2.3 Electrostatic potentials computed for two structurally different inhibitors of acetylcholinesterase computed using UCSF Chimera v1.11.2. Bound structures were obtained using flexible docking in Autodock Vina.
Figure 2.4 (a) Fukui indices, f⁺, computed for 4α,β-unsaturated carbonyls using Hirshfeld charges at the mPW1PW91/MIDIX+ level of theory. Maxima in the Fukui function are labeled with a black dot and a corresponding value; black circles mark the next highest values. Free energies of activation were calculated with the PM6 semiempirical method in the gas phase. Sensitization potential categories were derived from LLNA EC3% values [31]. (b) The LUMO orbital for 4.
Figure 2.5 Two-electron oxidation of hydroquinone (HQ) and t-butyl hydroquinone (tBHQ) to quinones, calculated at the M06-HF/6-31+G(d) level of theory in the gas phase. The solid black line represents energy difference between the HQ and tBHQ pathways, each recorded relative to enthalpies of the fully reduced HQ and tBHQ, respectively. The gray line represents the difference between the HQ pathway and the energetically less favorable tBHQ pathway. Each specie considered in the oxidation process is recorded below the graph with t-butyl substituents omitted for clarity. The species resulting from superoxide radical anion generation, phenoxy radical, and quinone are about 2.5 and 2.7 kcal/mol lower in energy in the (major) tBHQ than in the HQ pathway.
Figure 2.6 Linear models for free activation energies (a) and free energies of reaction (b) for nucleophilic substitutions of halides, epoxides, and tosylates; ΔG^† and ΔG_rxn values were computed in aqueous solution at the M06-2X/6-311 + G(d,p) level of theory; ΔG^≠ = 1706.38s_α − 27.26EE − 243.69S − 1.76SASA_α + 35.72(S × SASA_α) − 4.02; N = 15; R² = 0.98; $c02-math-064$ = 0.97; RMS = 0.96. ΔG_rxn = 801.01s_α − 4.12 μ + 8.90SASA_α + 2.04(μ × SASA_α) − 70.04; N = 15; R² = 0.95; $c02-math-065$ = 0.93; RMS = 0.35. s_β = local softness on the α carbon; EE = electrostatic solvation energy; S = global softness; SASA_α = surface accessible solvent area on the α carbon; μ = chemical potential [41].
Figure 2.7 (a) Active site of MIF (1CA7) with bound HPP in the keto form from QM/MM/MC simulations. (b)Truncated MIF–HPP complex with about 680 water molecules; the ligand is marked in black.
Figure 2.8 (a) Computed 2D free energy map for the H₂ proton transfer (see Figure 2.7). The white dashed line follows the minimum free energy path. (b) Snapshots of the transition state (TS) and the enolate intermediate illustrating relevant electrostatic interactions. The resolution based on a single FEP window is 0.025 Å.
Figure 2.9 Scatter plots of the octanol-water partition and distribution coefficients (log P and log D) versus HOMO-LUMO gap (Δϵ) calculated at the mPW1PW91/MIDIX+ level of theory. The 500+ compounds represented are colored by category of concern for acute aquatic toxicity (red = high concern; orange = medium concern; yellow = low concern; green = no concern) based on a 96-h toxicity assay of the fathead minnow [14]. The highlighted upper-left quadrant marks the “safer chemical space” (log P/D < 1.7; Δϵ > 6 eV), which should be targeted in designing new molecules.
Figure 2.10 Benzyloxazole molecule used in the FEP study by Cole et al. [54] and Bollini et al. [55]. The R group was iteratively modified to optimize binding affinity toward HIV-RT.

Chapter 3: Computational Approaches for Predicting hERG Activity

Figure 3.1 Structural representation of hERG channel generated through homology modeling. This model was generated using the open conformation of the hERG channel (UniProt accession number: Q12809) and the KvAP crystal structure (PDB code: 1ORQ) of Mus musculus [110] as template. The model was generated using a similar protocol reported by Farid et al. [102]. (a) Tetrameric representation of hERG channel. (b) Dimeric representation of S5 and S6 segments. The residues usually involved in drug interaction are represented by sticks. Each black sphere represents a potassium ion. (See color plate section for the color representation of this figure.)
Figure 3.2 Outcome interpretation from the Pred-hERG web app. Binary prediction, multiclass prediction, and predicted probability maps (PPM) extracted from binary models using Morgan fingerprints with 2048 bits. In the PPMs, green atoms or fragments represent contribution toward blockage of hERG, while pink indicate that it contributes to decrease of hERG blockage, and gray means no contribution. Gray isolines delimit the region of split between the positive (green) and the negative (pink) contribution. (See color plate section for the color representation of this figure.)
Figure 3.3 Comparison of structural alerts and the Pred-hERG QSAR models for prediction of hERG blockage. (a) Tertiary amines. (b) Aryl chloride. PP, predicted probability. (See color plate section for the color representation of this figure.)

Chapter 4: Computational Toxicology for Traditional Chinese Medicine

Figure 4.1 An OAT1 inhibitor pharmacophore model that consists of a negative ionizable feature (F1, red), one hydrophobe (F2, yellow), and a third feature that can be an aromatic center or a hydrophobic centroid (F3, yellow). In addition, six excluded volumes shown as gray spheres were present in this model. A potent OAT1 inhibitor, bumetanide (IC₅₀ = 6 μM), has been displayed with the model and the atoms are colored by atom type (carbon, gray; nitrogen, blue; oxygen, red; phosphorus, yellow).
Figure 4.2 The distribution of predicted TCM compounds with OAT1 inhibitory activity in medicinal TCMs. The black bars represent TCMs with three or more compounds mapped to the pharmacophore model and the names of these TCMs are listed in the figure.
Figure 4.3 TCM compounds mapping to the OAT1 inhibitor pharmacophore. The pharmacophore consists of a negative ionizable feature (red) and two hydrophobic features (yellow). For clarity, the excluded volumes are not shown here. (a) rhein; (b) aristolochic acid I; (c) salvianolic acid A; (d) lithospermic acid; (e) rosmarinic acid; (f) ferulic acid; (g) sinapinic acid; (h) and isoferulic acid.

Chapter 5: Pharmacophore Models for Toxicology Prediction

Figure 5.1 Visualization of exemplary pharmacophore models with selected chemical features in different modeling programs. Interactions of equilin cocrystallized with 17β-hydroxysteroid dehydrogenase 1 (PDB entry 1equ [5]) are shown. *, LigandScout; #, discovery studio; §, molecular operating system (MOE) [6]. The functionalities are abbreviated as H, hydrophobic; HBD, hydrogen bond donor; HBA, hydrogen bond acceptor; Xvol, exclusion volume.
Figure 5.2 PPARγ with bound tetrabromobisphenol A (PDB entry 3osw). The pharmacophore model derived from this endocrine disruptor–receptor complex can be refined and used for virtual screening for other PPARγ ligands from environmental chemicals. Protein–ligand interactions are color coded according to Figure 5.1.
Figure 5.3 Pharmacophore models for hERG blockers used in a parallel way. Each model covers a different fraction of active compounds, but is restrictive enough not to find a large number of inactive hits. All hit lists together cover the vast majority of active compounds and find less false positive hits compared to one very general model designed to cover most active compounds at once.
Figure 5.4 Protein–ligand interactions determined by X-ray crystallography of exemplary GPCRs. (a) β1-Adrenoceptor in complex with cyanopindolol (PDB entry 4bvn [51]); (b) dopamine D3 receptor in complex with eticlopride (PDB entry 3pbl [52]); (c) histamine H1 receptor in complex with doxepin (PDB entry 3rze [53]); (d) the protein–ligand interactions of all models superimposed onto each other; (e) all example models share two hydrophobic (yellow) features and a positively ionizable nitrogen (blue star). Figure inspired by Klabunde et al. [47].
Figure 5.5 11β-HSDs catalyze the interconversion of the active glucocorticoid cortisone and its inactive metabolite cortisol [6].
Figure 5.6 Interconversion of sex hormones and their metabolites catalyzed by 17β-HSDs [6].
Figure 5.7 Structure–activity relationship rationalization of chemical UV filters of the benzophenone class-inhibiting 17β-HSD3. The three hydrogen bond acceptors (red) and two aromatic rings (blue) are essential for bioactivity. Etherification of one of the hydroxyl groups inactivates the compound (arrows). *, Residual activity of the enzyme was measured at a compound concentration of 20 μM [6].

Chapter 6: Transporters in Hepatotoxicity

Figure 6.1 Transporters located in the hepatocyte. The medium grey symbols represent the canalicular transporters and dark grey ones the basolateral transporters. Cycles represent uptake transporters and ellipses refer to efflux transporters. The arrows define the direction of transport.
Figure 6.2 The cycle of bilirubin in the liver. Bilirubin is taken up from sinusoidal blood by OATP1B1 and OATP1B3. It is metabolized by UGT1A1 into mono- and bi-glucuronidated products that are exported into bile primarily by MRP2 and in smaller extent (smaller arrow) by BCRP. A portion of the glucuronidated or unglucuronidated bilirubin is effluxed into sinusoidal blood by MRP4 and the cycle is repeated.

Chapter 7: Cheminformatics in a Clinical Setting

Figure 7.1 Illustration of structural similarity. Using phencyclidine (PCP) as the target compound, 2D similarity to five different compounds was calculated using MDL public keys and the Tanimoto coefficient; three of these (dextromethorphan, chlorpromazine and tramadol) have been reported to cross-react with at least some marketed PCP immunoassays, and the other two (ketamine and ibuprofen) have not been reported to cross-react with PCP screening assays. PCP has the highest similarity (in descending order) to dextromethorphan, chlorpromazine, and tramadol. PCP has low structural similarity to ketamine (despite having similar pharmacological properties to PCP) and essentially no structural similarity to ibuprofen.
Figure 7.2 Similarity of drugs and drug metabolites relative to the target compounds for four broadly specific DOA/Tox assays. Cross-reactivity data for four DOA/Tox assays were sorted into six categories. The similarity (using MDL public keys and the Tanimoto coefficient) of each tested compound to the target compound of the DOA/Tox assay was plotted. (a) Amphetamine assays (using d-amphetamine as the target). (b) Barbiturate assays (using secobarbital as the target compound). (c) Benzodiazepine assays (using diazepam as the target compound). (d) TCA assays (using desipramine as the target compound). (
Figure 7.3 Similarity of drugs and drug metabolites relative to the target compounds for four DOA/Tox assays. Cross-reactivity data for four DOA/Tox assays were sorted into six categories. The similarity (using MDL public keys and the Tanimoto coefficient) of each tested compound to the target compound of the DOA/Tox assay of the DOA/Tox assay was plotted. (a) Cannabinoid assays (using 9-carboxy-11-nor-Δ₉-tetrahydrocannabinol as the target compound). (b) Cocaine metabolite (benzoylecgonine) assays. (c) Opiate assays (using morphine as the target compound). (d) Phencyclidine assays.
Figure 7.4 Similarity of drugs and drug metabolites relative to the target compounds for four TDM assays. Cross-reactivity data for four TDM assays were sorted into three categories. The similarity (using MDL public keys and the Tanimoto coefficient) of each tested compound to the target compound of each assay was plotted. (a) Cyclosporine assays. (b) Lamotrigine assays. (c) Theophylline assays. (d) Valproic acid assays.
Figure 7.5 Cortisol immunoassay cross-reactivity and similarity predictions. (a) The plot shows the cortisol reference range for adults (highlighted in yellow) in comparison to the predicted apparent cortisol concentrations produced on the Roche Elecsys Cortisol assay by 6-methylprednisolone, prednisolone, 21-deoxycortisol (healthy controls and patients with 21-hydroxylase deficiency), and 11-deoxycortisol (healthy controls, patients with 11β-hydroxylase deficiency, and following metyrapone challenge). Table 1 contains the concentration ranges and percentage of cross-reactivity values from which the estimated apparent cortisol concentrations are derived. (b) Two-dimensional similarity of compounds to cortisol is shown, sorted by degree of cross-reactivity in the Roche Cortisol assay (horizontal line in each column indicates average similarity within that group). Similarity values vary from 0 to 1, with 1 being maximally similar. The compounds are subdivided into categories of strong cross-reactivity (5% or greater, black circles), weak cross-reactivity (0.5-4.9%, red squares), very weak cross-reactivity (0.05-0.49%, blue triangles), and no cross-reactivity (<0.05%, green diamonds) to the Roche Cortisol assay.
Figure 7.6 Testosterone immunoassay cross-reactivity and similarity predictions. (a) The plot shows the testosterone reference range for males and females (highlighted in yellow) in comparison to the predicted apparent testosterone concentrations produced on the Roche Elecsys Testosterone II assay by methyltestosterone, norethindrone, nandrolone, and androstenedione (healthy controls and patients with 21-hydroxylase deficiency). Table 5 contains the concentration ranges and percent cross-reactivity values from which the estimated apparent testosterone concentrations are derived. (b) Two-dimensional similarity of compounds to testosterone is shown, sorted by degree of cross-reactivity in the Roche Testosterone II assay (horizontal line in each column indicates average similarity within that group). Similarity values vary from 0 to 1, with 1 being maximally similar. The compounds are subdivided into categories of strong cross-reactivity (5% or greater, black circles), weak cross-reactivity (0.5-4.9%, red squares), very weak cross-reactivity (0.05-0.49%, blue triangles), and no cross-reactivity (<0.05%, green diamonds) to the Roche Testosterone II assay.
Figure 7.7 Representative chemical structures of amphetamine-like drugs. Abbreviations: MDMA, 3,4-methylenedioxy-N-methamphetamine; MDPV, 3,4-methylenedioxypyrovalerone; MDPBP, 3′,4′-methylenedioxy-α-pyrrolidinobutiophenone.
Figure 7.8 Representative chemical structures of cannabinoids.
Figure 7.9 Similarity comparisons of cannabinoids. Test compounds are divided into broad categories (JWH series, AM/UR/RCS/XLR series, other synthetic cannabinoids not possessing the chemical backbone of THC, cannabinoids sharing chemical backbone of THC, endogenous eicosanoid cannabinoids, and non-cannabinoids). (a) 2D similarity of the N-pentanoic acid metabolite of JWH-018 to 168 other compounds. (b) 2D similarity of the N-butanoic acid metabolite of JWH-073 to 168 other compounds. (c) 2D similarity of the N-4-hydroxy metabolite of JWH-250 to 168 other compounds. (d) 2D similarity of 9-carboxy-THC to 168 other compounds.

Chapter 8: Computational Tools for ADMET Profiling

Figure 8.1 Predictive QSAR modeling workflow [38].
Figure 8.2 Chemical and biological similarities do not correlate. Each point represents a pair of compounds characterized by pairwise chemical similarity (y-axis) versus biological similarity (x-axis). Examples of a priori outliers that should be flagged and analyzed separately are shown.
Figure 8.3 Strategies for utilizing diverse data streams for predicting higher-order biological effects.
Figure 8.4 Two-step hierarchical k-nearest neighbor (kNN) QSAR workflow to develop an enhanced rat acute toxicity (LD₅₀) model by using cytotoxicity data (IC₅₀) as biological profile descriptors.
Figure 8.5 Main window of the CBRA program divided into three parts: (a) selection of input files, (b) colored radial plot, (c) molecular structure viewer, textual information, and options.
Figure 8.6 Graphical representation of a local neighborhood for Benoxacor [65].

Chapter 9: Computational Toxicology and Reach

Figure 9.1 The first page of the output of the CAESAR model for the target compound.
Figure 9.2 The page with the confidence interval.
Figure 9.3 The list of the sixth most similar compounds.
Figure 9.4 The applicability domain index and its components.
Figure 9.5 The chart with the Mlog P and logBCF plots.
Figure 9.6 The panel to access information about the models.
Figure 9.7 Estimations provided by the four individual models plus the consensus for a chemical used as example (ethyl 2-bromobutanoate).
Figure 9.8 The six most similar compounds to the target present in the original dataset and their respective observed and predicted values.
Figure 9.9 The SA8 and the three most similar compounds of the training set with the same SA.
Figure 9.10 The SA SM93 and the three most similar compounds of the training set with the same SA.

Chapter 10: Computational Approaches to Predicting Dermal Absorption of Complex Topical Mixtures

Figure 10.1 Light micrograph of normal human skin. SC, stratum corneum; E, epidermis; D, Dermis. —— = 50 µm.
Figure 10.2 Relationship between anatomical regions in skin where chemical mixture modulation of absorption can occur, physicochemical processes involved, and type of modeling employed.
Figure 10.3 Relationship between a QSAR model without and with a mixture factor component included. Note that the overall slope of the QSAR is dependent upon the penetrant's chemical properties, while mixture component effects result in columns along the penetrant property unless mixture components based on the mixture properties are also included.
Figure 10.4 QSPR model fit of porcine skin diffusion cell data using diffusivity (D) and partition (K) coefficients estimated by a random process dermatokinetic model [59].
Figure 10.5 QSPR model fit of permeability coefficients obtained from a dermatokinetic model of in situ perfused porcine skin flap [60].
Figure 10.6 Compartmental pharmacokinetic model linking skin absorption determined in an in vitro model to a systemic model to predict plasma concentration time profiles in vivo.
Figure 10.7 A comparison model fits with (Model 2006) and without (Model 1005) incorporation of a random process to account for transient changes in diffusivity [59].

Chapter 11: Big Data in Computational Toxicology: Challenges and Opportunities

Figure 11.1 The “four V's” of big data can be used to describe the properties of these fast growing chemical toxicity data.
Figure 11.2 Increase in the number of compounds and bioassays recorded in PubChem within eight years (from September 2008 to September 2015).

Chapter 12: HLA-Mediated Adverse Drug Reactions: Challenges and Opportunities for Predictive Molecular Modeling

Figure 12.1 Structure of HLA-variants Class I and Class II.
Figure 12.2 HLA-drug binding mechanism adapted from Illing et al. [48] for T-cell activation (a) Altered repertoire (non-covalent). (B) p. i. complex (non-covalent). (C) Hapten complex (non-covalent). The non-covalent T-cell interaction is not shown.
Figure 12.3 Alternative p. i. complex CD8⁺ T-cell signaling pathway for carbamazepine binding with HLA-B*15:02.
Figure 12.4 Schematic for using molecular docking to perform virtual screening at HLA-B*57:01 variant.
Figure 12.5 (a) Self-docking of abacavir alignment, (b) binding modes of abacavir, (c), (d) docking results.
Figure 12.6 Chemical structures of the 14 drugs used to construct a virtual screening molecular docking model.
Figure 12.7 Heat map of DS for full set. Green spaces represent the most favorable docking scores (DS < −7 kcal/mol), while spaces transition from orange to red represent drugs that have nonfavorable interactions with HLA-B*57:01 (DS > −7 kcal/mol). White spaces indicate that GLIDE was unable to identify a best binding mode between drug and HLA-B*57:01.
Figure 12.8 Heat map of eM scores for the full test set. Green spaces represent the most favorable docking scores (eM < −50 kcal/mol) while spaces transition from yellow to red represent drugs that have nonfavorable interactions with HLA-B*57:01 (eM > −50 kcal/mol). White spaces indicate that GLIDE was unable to identify a best binding mode between drug and HLA-B*57:01.
Figure 12.9 Full docking summary combining SP and XP results. Green represents compounds that passed both DS (DS < −7 kcal/mol) and eM (eM < −50 kcal/mol) thresholds for SP and XP scoring functions; yellow represents compounds that passed the thresholds for XP but failed using SP; orange represents compounds that failed the XP thresholds but passed SP; and red represents the compounds that failed the thresholds for both XP and SP scoring functions.
Figure 12.10 Population distribution by percentage for three select HLA-variants: HLA-A*31:01, HLA-B*15:02, and HLA-B*57:01. The ethnicities studied were African-American (n = 251), Caucasian (n = 265), Hispanic (n = 234), North American Natives (n = 187), and Asians (n = 358) from the United States. Please refer to Cao et al. 2001 [73] for further details.

Chapter 13: Open Science Data Repository for Toxicology

Figure 13.1 (1) Examples of the OSDR prototype to date showing bidirectional integration with various cloud drives allows seamless data transfers between cloud storage and OSDR; (2) web user interface also allows intuitive data deposition using drag and drop; (3) concise filter system provides a convenient way of navigating information stored or indexed in OSDR; (4) hierarchical presentation of information allows one to arrange the data based on organization or research structure; (5) standard CMS (content management system) operations are supported; (6) various view modes allow representing complex information in a visual and concise manner; (7) user interface based on modern web frameworks provides an excellent user experience.
Figure 13.2 Examples of the OSDR prototype to date showing OSDR tabular data entry. Mapping columns from a CSV file (1) to their semantic meaning (2) allows to resolve entries in real-time into a set of public database identifiers (3, ChemSpider, ChEMBL, PubChem), create a chemical structure from provided information (4), and calculate conversion confidence value based on a set of mappings (e.g., chemical name, InChI, SMILES).
Figure 13.4 Examples of the OSDR prototype to date. Built in preview mode showing different file types (1, word; 2, excel; 3, powerpoint; 4, PDF).
Figure 13.5 OSDR microservice overview.
Figure 13.3 Examples of the OSDR prototype to date. (a) Document browse mode with thumbnail previews. (b) Document view mode with a larger preview and other information arranged into infoboxes.
Figure 13.6 Logical architecture of OSDR with cheminformatics and machine learning modules.
Figure 13.7 OSDR microservice-oriented architecture.
Figure 13.8 OSDR development workflow.

Chapter 14: Developing Next Generation Tools for Computational Toxicology

Figure 14.1 Screenshots of the Mobile Molecular DataSheet.
Figure 14.2 The Green Solvents app. (a) Molecule overview. (b) Molecule details list scores (good = 1, bad = 10) for safety, health, flammability, environment, waste, reactivity, and lifecycle criteria. (See color plate section for the color representation of this figure.)
Figure 14.3 Examples of green reactions from the Green Lab Notebook app.
Figure 14.4 Example of preliminary work. (a) Highlighting molecules using Bayesian models for various ADME/Tox properties. (b) Clustering molecules using fingerprint descriptors. (See color plate section for the color representation of this figure.)
Figure 14.5 Visualization of data cut-offs, ROC plots, and active and inactive molecules for hERG Ki data from ChEMBL.
Figure 14.6 Preliminary work using open datasets and computed models for (a) EPA Tox21 data used to make predictions that are visualized in the PolyPharma mobile app. (b) Novel visualization and prediction methods in PolyPharma showing atom highlighting for each model and clustering. http://itunes.apple.com/app/polypharma/id1025327772. (See color plate section for the color representation of this figure.)
Figure 14.7 Assay Central schematic.

List of Tables

Chapter 1: Accessible Machine Learning Approaches for Toxicology

Table 1.1 Comparison of machine learning methods using FCFP6 1024 bit descriptors on ADME/Tox properties using fivefold cross-validation ROC values
Table 1.2 Comparison of machine learning methods using FCFP6 1024-bit descriptors on ADME/Tox properties using fivefold cross-validation F1 values at p = 0.5

Chapter 2: Quantum Mechanics Approaches in Computational Toxicology

Table 2.1 Determining selected physicochemical properties for a clinical prodrug (CAS 623152-11-4) from an MC simulation versus a simple geometry optimization using the AM1 semiempirical method
Table 2.2 Examples of global electronic parameters calculated from frontier molecular orbitals
Table 2.3 Global electronic parameters for 3-penten-2-one, propargyl acrylate, allyl acrylate, and methyl acrylate derived from HOMO and LUMO energies at the mPW1PW91/MIDIX + level of theory
Table 2.4 Local electronic parameters derived from density functional theory
Table 2.5 Hydrogen bonding (HB) as an inverse metric of human skin permeability
Table 2.6 Reaction barriers correlated to GSH-binding rate constants for methylacrolein, 3-penten-2-one, allyl acrylate, and ethyl crotonate
Table 2.7 Reaction energetics as a predictor of mutagenicity potentials for 2,2-difluorooxirane, 2,2-dichlorooxirane, 2,3-dichlorooxirane, and 2,2,3-trichlorooxirane
Table 2.8 Reaction energetics computed for three benzyl halide skin sensitizers at the M06-2X/6-311 + G(d, p) level in gas phase and in aqueous solution

Chapter 3: Computational Approaches for Predicting hERG Activity

Table 3.1 QSAR studies for predicting hERG blockage, published during the period 2014-2016

Chapter 4: Computational Toxicology for Traditional Chinese Medicine

Table 4.1 Representative molecules used for OAT1 inhibitor pharmacophore model generation and validation
Table 4.2 Example TCM compounds with experimental information about interactions with OAT1.
Table 4.3 Structurally similar TCM compounds without experimental validation

Chapter 5: Pharmacophore Models for Toxicology Prediction

Table 5.1 Experimentally validated pharmacophore models for 17β-HSD inhibitors

Chapter 6: Transporters in Hepatotoxicity

Table 6.1 Summary of the best-performing models for transporters

Chapter 9: Computational Toxicology and Reach

Table 9.1 The main differences between old models and models for regulatory purposes

Chapter 10: Computational Approaches to Predicting Dermal Absorption of Complex Topical Mixtures

Table 10.1 Experimental variables that should be controlled or documented when conducting dermal absorption studies

Chapter 11: Big Data in Computational Toxicology: Challenges and Opportunities

Table 11.1 Public available toxicity data resources (as of 10/22/2016)
Table 11.2 Twenty human toxicants with their relevant PubChem bioassay responses

Chapter 12: HLA-Mediated Adverse Drug Reactions: Challenges and Opportunities for Predictive Molecular Modeling

Table 12.1 List of drug-HLA associations with their reported odds ratios

Chapter 14: Developing Next Generation Tools for Computational Toxicology

Table 14.1 Mobile apps for chemistry developed by Molecular Materials Informatics, Inc

Wiley Series on Technologies for the Pharmaceutical Industry Sean Ekins, Series Editor

Computational Toxicology: Risk Assessment for Pharmaceutical and Environmental Chemicals

Edited by Sean Ekins

Pharmaceutical Applications of Raman Spectroscopy

Edited by Slobodan Šašić

Pathway Analysis for Drug Discovery: Computational Infrastructure and Applications

Edited by Anton Yuryev

Drug Efficacy, Safety, and Biologics Discovery: Enmerging Technologies and Tools

Edited by Sean Ekins and Jinghai J. Xu

The Engines of Hippocrates: From the Dawn of Medicine to Medical and Pharmaceutical Informatics

Barry Robson and O.K. Baek

Pharmaceutical Data Mining: Applications for Drug Discovery

Edited by Konstantin V. Balakin

The Agile Approach to Adaptive research: Optimizing Efficiency in Clinical Development

Michael J. Rosenberg

Pharmaceutical and Biomedical Project Management in a Changing Global Environment

Edited by Scott D. Babler

Systems Biology in Drug Discovery and Development

Edited by Daniel L. Young and Seth Michelson

Collaborative Computational Technologies for Biomedical Research

Edited by Sean Ekins, Maggie A.Z. Hupcey and Antony J. William

Predictive Approaches in Drug Discovery and Development: Biomarkers and In Vitro/ In Vivo correlations

Edited by J. Andrew Williams, Richard Lalonde, Jeffrey Koup and David D. Christ

Collaborative Innovation in Drug Discovery, Strategies for Public and Private Partnerships

Edited by Rathnam Chaguturu

Computational Toxicology: Risk Assessment for Chemicals

Edited by Sean Ekins

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by law. Advice on how to obtain permission to reuse material from this title is available at http://www.wiley.com/go/permissions.

The right of Sean Ekins to be identified as the Editor in this work has been asserted in accordance with law.

For details of our global editorial offices, customer services, and more information about Wiley products visit us at www.wiley.com.

Wiley also publishes its books in a variety of electronic formats and by print-on-demand. Some content that appears in standard print versions of this book may not be available in other formats.

In view of ongoing research, equipment modifications, changes in governmental regulations, and the constant flow of information relating to the use of experimental reagents, equipment, and devices, the reader is urged to review and evaluate the information provided in the package insert or instructions for each chemical, piece of equipment, reagent, or device for, among other things, any changes in the instructions or indication of usage and for added warnings and precautions. While the publisher and authors have used their best efforts in preparing this work, they make no representations or warranties with respect to the accuracy or completeness of the contents of this work and specifically disclaim all warranties, including without limitation any implied warranties of merchantability or fitness for a particular purpose. No warranty may be created or extended by sales representatives, written sales materials or promotional statements for this work. The fact that an organization, website, or product is referred to in this work as a citation and/or potential source of further information does not mean that the publisher and authors endorse the information or services the organization, website, or product may provide or recommendations it may make. This work is sold with the understanding that the publisher is not engaged in rendering professional services. The advice and strategies contained herein may not be suitable for your situation. You should consult with a specialist where appropriate. Further, readers should be aware that websites listed in this work may have changed or disappeared between when this work was written and when it is read. Neither the publisher nor authors shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages.

Title: Computational toxicology : risk assessment for chemicals / edited by Sean Ekins.

Description: First edition. | Hoboken, NJ : John Wiley & Sons, 2018. | Series: Wiley series on technologies for the pharmaceutical industry | Includes bibliographical references and index. |

Identifiers: LCCN 2017037714 (print) | LCCN 2017046458 (ebook) | ISBN 9781119282570 (pdf) | ISBN 9781119282587 (epub) | ISBN 9781119282563 (cloth)

Subjects: LCSH: Toxicology-Mathematical models. | Toxicology-Computer simulation. | QSAR (Biochemistry)

Classification: LCC RA1199.4.M37 (ebook) | LCC RA1199.4.M37 C66 2018 (print) | DDC 615.90285-dc23

I should have no objection to go over the same life from its beginning to the end: requesting only the advantage authors have, of correcting in a second edition the faults of the first.

List of Contributors

Ni Ai

Pharmaceutical Informatics Institute

College of Pharmaceutical Sciences

Zhejiang University

Hangzhou

Zhejiang, PR

China

Vinicius M. Alves

LabMol – Laboratory for Molecular Modeling and Design, Faculty of Pharmacy

Federal University of Goias

Goiania, GO

Brazil

Carolina Horta Andrade

LabMol – Laboratory for Molecular Modeling and Design, Faculty of Pharmacy

Federal University of Goias

Goiania, GO

Brazil

Rodolpho C. Braga

LabMol – Laboratory for Molecular Modeling and Design, Faculty of Pharmacy

Federal University of Goias

Goiania, GO

Brazil

Jason Chittenden

Center for Chemical Toxicology Research and Pharmacokinetics Biomathematics Program

North Carolina State University

Raleigh, NC

USA

Alex M. Clark

Molecular Materials Informatics, Inc.

Montreal, Quebec

Canada

Daniela Digles

Department of Pharmaceutical Chemistry

University of Vienna

Wien

Austria

George van Den Driessche

Department of Chemistry

Bioinformatics Research Center

North Carolina State University

Raleigh, NC

USA

Gerhard F. Ecker

Department of Pharmaceutical Chemistry

University of Vienna

Wien

Austria

Sean Ekins

Collaborations Pharmaceuticals, Inc.

Raleigh, NC

USA

Emilio Benfenati

IRCCS – Istituto di Ricerche Farmacologiche “Mario Negri”

Laboratory of Environmental Chemistry and Toxicology

Milan

Italy

Xiaohui Fan

Pharmaceutical Informatics Institute

College of Pharmaceutical Sciences

Zhejiang University

Hangzhou

Zhejiang, PR

China

Denis Fourches

Department of Chemistry

Bioinformatics Research Center

North Carolina State University

Raleigh, NC

USA

Joel S. Freundlich

Department of Pharmacology & Physiology

New Jersey Medical School

Rutgers University

Newark, NJ

USA

and

Division of Infectious Disease

Department of Medicine and the Ruy V. Lourenço Center for the Study of Emerging and Re-emerging Pathogens

New Jersey Medical School, Rutgers University

Newark, NJ

USA

Chris Grulke

National Center for Computational Toxicology, Office of Research and Development

U.S. Environmental Protection Agency

Research Triangle Park

Durham, NC

USA

Sankalp Jain

Department of Pharmaceutical Chemistry

University of Vienna

Wien

Austria

Alexandru Korotcov

Gaithersburg, MD

USA

Jakub Kostal

Chemistry Department

The George Washington University

Washington DC

USA

Eleni Kotsampasakou

Department of Pharmaceutical Chemistry

University of Vienna

Wien

Austria

Matthew D. Krasowski

Department of Pathology

University of Iowa Hospitals and Clinics

Iowa City, IA

USA

Mary A. Lingerfelt

Collaborations Pharmaceuticals, Inc.

Raleigh, NC

USA

Anna Lombardo

IRCCS – Istituto di Ricerche Farmacologiche “Mario Negri”

Laboratory of Environmental Chemistry and Toxicology

Milan

Italy

Grace Patlewicz

National Center for Computational Toxicology, Office of Research and Development

U.S. Environmental Protection Agency

Research Triangle Park

Durham, NC

USA

Alexander L. Perryman

Department of Pharmacology & Physiology

New Jersey Medical School

Rutgers University

Newark, NJ

USA

Ann Richard

National Center for Computational Toxicology, Office of Research and Development

U.S. Environmental Protection Agency

Research Triangle Park

Durham, NC

USA

Jim E. Riviere

Center for Chemical Toxicology Research and Pharmacokinetics Biomathematics Program

North Carolina State University

Raleigh, NC

USA

Alessandra Roncaglioni

IRCCS – Istituto di Ricerche Farmacologiche “Mario Negri”

Laboratory of Environmental Chemistry and Toxicology

Milan

Italy

Daniela Schuster

Institute of Pharmacy/Pharmaceutical Chemistry

University of Innsbruck

Innsbruck

Austria

Imran Shah

National Center for Computational Toxicology, Office of Research and Development

U.S. Environmental Protection Agency

Research Triangle Park

Durham, NC

USA

Valery Tkachenko

Rockville, MD

USA

Alexander Tropsha

UNC Eshelman School of Pharmacy

University of North Carolina at Chapel Hill

Chapel Hill, NC

USA

John Wambaugh

National Center for Computational Toxicology, Office of Research and Development

U.S. Environmental Protection Agency

Research Triangle Park

Durham, NC

USA

Antony J. Williams

National Center for Computational Toxicology, Office of Research and Development

U.S. Environmental Protection Agency

Research Triangle Park

Durham, NC

USA

Richard Zakharov

Rockville, MD

USA

Linlin Zhao

Center for Computational and Integrative Biology

Rutgers University

Camden, NJ

USA

Hao Zhu

Center for Computational and Integrative Biology

Rutgers University

Camden, NJ

USA

and

Department of Chemistry

Rutgers University

Camden, NJ

USA

Kimberley M. Zorn

Collaborations Pharmaceuticals, Inc.

Raleigh, NC

USA