Saagar-A New, Extensible Set of Molecular Substructures for QSAR/QSPR and Read-Across Predictions.

Molecular structure-based predictive models provide a proven alternative to costly and inefficient animal testing. However, due to a lack of interpretability of predictive models built with abstract molecular descriptors they have earned the notoriety of being black boxes. Interpretable models require interpretable descriptors to provide chemistry-backed predictive reasoning and facilitate intelligent molecular design. We developed a novel set of extensible chemistry-aware substructures, Saagar, to support interpretable predictive models and read-across protocols. Performance of Saagar in chemical characterization and search for structurally similar actives for read-across applications was compared with four publicly available fingerprint sets (MACCS (166), PubChem (881), ECFP4 (1024), ToxPrint (729)) in three benchmark sets (MUV, ULS, and Tox21) spanning ∼145 000 compounds and 78 molecular targets at 1%, 2%, 5%, and 10% false discovery rates. In 18 of the 20 comparisons, interpretable Saagar features performed better than the publicly available, but less interpretable and fixed-bit length, fingerprints. Examples are provided to show the enhanced capability of Saagar in extracting compounds with higher scaffold similarity. Saagar features are interpretable and efficiently characterize diverse chemical collections, thus making them a better choice for building interpretable predictive in silico models and read-across protocols.

An evaluation of in-house and off-the-shelf in silico models: implications on guidance for mutagenicity assessment.

The evaluation of impurities for genotoxicity using in silico models are commonplace and have become accepted by regulatory agencies. Recently, the ICH M7 Step 4 guidance was published and requires two complementary models for genotoxicity assessments. Over the last ten years, many companies have developed their own internal genotoxicity models built using both public and in-house chemical structures and bacterial mutagenicity data. However, the proprietary nature of internal structures prevents sharing of data and the full OECD compliance of such models. This analysis investigated whether using in-house internal compounds for training models is needed and substantially impacts the results of in silico genotoxicity assessments, or whether using commercial-off-the-shelf (COTS) packages such as Derek Nexus or Leadscope provide adequate performance. We demonstrated that supplementation of COTS packages with a Support Vector Machine (SVM) QSAR model trained on combined in-house and public data does, in fact, improve coverage and accuracy, and reduces the number of compounds needing experimental assessment, i.e., the liability load. This result indicates that there is added value in models trained on both internal and public structures and incorporating such models as part of a consensus approach improves the overall evaluation. Lastly, we optimized an in silico consensus decision-making approach utilizing two COTS models and an internal (SVM) model to minimize false negatives.

Quantitative structure-activity relationship models of clinical pharmacokinetics: clearance and volume of distribution.

Reliable prediction of two fundamental human pharmacokinetic (PK) parameters, systemic clearance (CL) and apparent volume of distribution (Vd), determine the size and frequency of drug dosing and are at the heart of drug discovery and development. Traditionally, estimated CL and Vd are derived from preclinical in vitro and in vivo absorption, distribution, metabolism, and excretion (ADME) measurements. In this paper, we report quantitative structure-activity relationship (QSAR) models for prediction of systemic CL and steady-state Vd (Vdss) from intravenous (iv) dosing in humans. These QSAR models avoid uncertainty associated with preclinical-to-clinical extrapolation and require two-dimensional structure drawing as the sole input. The clean, uniform training sets for these models were derived from the compilation published by Obach et al. (Drug Metab. Disp. 2008, 36, 1385-1405). Models for CL and Vdss were developed using both a support vector regression (SVR) method and a multiple linear regression (MLR) method. The SVR models employ a minimum of 2048-bit fingerprints developed in-house as structure quantifiers. The MLR models, on the other hand, are based on information-rich electro-topological states of two-atom fragments as descriptors and afford reverse QSAR (RQSAR) analysis to help model-guided, in silico modulation of structures for desired CL and Vdss. The capability of the models to predict iv CL and Vdss with acceptable accuracy was established by randomly splitting data into training and test sets. On average, for both CL and Vdss, 75% of test compounds were predicted within 2.5-fold of the value observed and 90% of test compounds were within 5.0-fold of the value observed. The performance of the final models developed from 525 compounds for CL and 569 compounds for Vdss was evaluated on an external set of 56 compounds. The predictions were either better or comparable to those predicted by other in silico models reported in the literature. To demonstrate the practical application of the RQSAR approach, the structure of vildagliptin, a high-CL and a high-Vdss compound, is modified based on the atomic contributions to its predicted CL and Vdss to propose compounds with lower CL and lower Vdss.

Metabolic-intermediate complex formation with cytochrome P450: theoretical studies in elucidating the reaction pathway for the generation of reactive nitroso intermediate.

Mechanism-based inhibition (MBI) of cytochrome P450 (CYP) can lead to drug-drug interactions and often to toxicity. Some aliphatic and aromatic amines can undergo biotransformation reactions to form reactive metabolites such as nitrosoalkanes, leading to MBI of CYPs. It has been proposed that the nitrosoalkanes coordinate with the heme iron, forming metabolic-intermediate complex (MIC), resulting in the quasi-irreversible inhibition of CYPs. Limited mechanistic details regarding the formation of reactive nitroso intermediate and its coordination with heme-iron have been reported. A quantum chemical analysis was performed to elucidate potential reaction pathways for the generation of nitroso intermediate and the formation of MIC. Elucidation of the energy profile along the reaction path, identification of three-dimensional structures of reactive intermediates and transition states, as well as charge and spin density analyses, were performed using the density functional B3LYP method. The study was performed using Cpd I [iron (IV-oxo] heme porphine with SH(-) as the axial ligand) to represent the catalytic domain of CYP, simulating the biotransformation process. Three pathways: (i) N-oxidation followed by proton shuttle, (ii) N-oxidation followed by 1,2-H shift, and (iii) H-abstraction followed by rebound mechanism, were studied. It was observed that the proton shuttle pathway was more favorable over the whole reaction leading to reactive nitroso intermediate. This study revealed that the MIC formation from a primary amine is a favorable exothermic process, involving eight different steps and preferably takes place on the doublet spin surface of Cpd I. The rate-determining step was identified to be the first N-oxidation of primary amine.

Assessing the toxic effects of ethylene glycol ethers using Quantitative Structure Toxicity Relationship models.

Experimental determination of toxicity profiles consumes a great deal of time, money, and other resources. Consequently, businesses, societies, and regulators strive for reliable alternatives such as Quantitative Structure Toxicity Relationship (QSTR) models to fill gaps in toxicity profiles of compounds of concern to human health. The use of glycol ethers and their health effects have recently attracted the attention of international organizations such as the World Health Organization (WHO). The board members of Concise International Chemical Assessment Documents (CICAD) recently identified inadequate testing as well as gaps in toxicity profiles of ethylene glycol mono-n-alkyl ethers (EGEs). The CICAD board requested the ATSDR Computational Toxicology and Methods Development Laboratory to conduct QSTR assessments of certain specific toxicity endpoints for these chemicals. In order to evaluate the potential health effects of EGEs, CICAD proposed a critical QSTR analysis of the mutagenicity, carcinogenicity, and developmental effects of EGEs and other selected chemicals. We report here results of the application of QSTRs to assess rodent carcinogenicity, mutagenicity, and developmental toxicity of four EGEs: 2-methoxyethanol, 2-ethoxyethanol, 2-propoxyethanol, and 2-butoxyethanol and their metabolites. Neither mutagenicity nor carcinogenicity is indicated for the parent compounds, but these compounds are predicted to be developmental toxicants. The predicted toxicity effects were subjected to reverse QSTR (rQSTR) analysis to identify structural attributes that may be the main drivers of the developmental toxicity potential of these compounds.

In silico approaches to predicting cancer potency for risk assessment of genotoxic impurities in drug substances.

The current risk assessment approach for addressing the safety of very small concentrations of genotoxic impurities (GTIs) in drug substances is the threshold of toxicological concern (TTC). The TTC is based on several conservative assumptions because of the uncertainty associated with deriving an excess cancer risk when no carcinogenicity data are available for the impurity. It is a default approach derived from a distribution of carcinogens and does not take into account the properties of a specific chemical. The purpose of the study was to use in silico tools to predict the cancer potency (TD(50)) of a compound based on its structure. Structure activity relationship (SAR) models (classification/regression) were developed from the carcinogenicity potency database using MultiCASE and VISDOM. The MultiCASE classification models allowed the prediction of carcinogenic potency class, while the VISDOM regression models predicted a numerical TD(50). A step-wise approach is proposed to calculate predicted numerical TD(50) values for compounds categorized as not potent. This approach for non-potent compounds can be used to establish safe levels greater than the TTC for GTIs in a drug substance.

Modeling ADMET.

Drug discovery and development is a costly and time-consuming endeavor (Calcoen et al. Nat Rev Drug Discov 14(3):161-162, 2015; The truly staggering cost of inventing new drugs. Forbes. http://www.forbes.com/sites/matthewherper/2012/02/10/the-truly-staggering-cost-of-inventing-new-drugs/, 2012; Scannell et al. Nat Rev Drug Discov 11(3):191-200, 2012). Over the last two decades, computational tools and in silico models to predict ADMET (Adsorption, Distribution, Metabolism, Excretion, and Toxicity) profiles of molecules have been incorporated into the drug discovery process mainly in an effort to avoid late-stage failures due to poor pharmacokinetics and toxicity. It is now widely recognized that ADMET issues should be addressed as early as possible in drug discovery. Here, we describe in detail how ADMET models can be developed and applied using a commercially available package, ADMET Predictor™ 7.2 (ADMET Predictor v7.2. Simulations Plus, Inc., Lancaster, CA, USA).

Role of ADME characteristics in drug discovery and their in silico evaluation: in silico screening of chemicals for their metabolic stability.

Drug discovery is a long, arduous process broadly grouped into disease target identification, target validation, high-throughput identification of "hits" and "leads", lead optimization, and pre-clinical and clinical evaluation. Each area is a vast discipline in itself. However, all but the first two stages involve, to varying degrees, the characterization of absorption, distribution, metabolism, excretion, (ADME), and toxicity (T) of the molecules being pursued as potential drug candidates. Clinical failures of about 50% of the Investigational New Drug (IND) filings are attributed to their inadequate ADMET attributes. It is, therefore, no surprise that, in the current climate of social and regulatory pressure on healthcare costs, the pharmaceutical industry is searching for any means to minimize this attrition. Building mathematical models, called in silico screens, to reliably predict ADMET attributes solely from molecular structure is at the heart of this effort in reducing costs as well as development cycle times. This article reviews the emerging field of in silico evaluation of ADME characteristics. For different approaches that have been employed in this area, a critique of the scope and limitations of their descriptors, statistical methods, and reliability are presented. For instance, are geometry-based descriptors absolutely essential or is lower-level structure quantification equally good? What advantages, if any, do we have for methods like artificial neural networks over the least squares optimization methods with rigorous statistical diagnostics? Is any in silico screen worth application, let alone interpretation, if it is not adequately validated? Once deemed acceptable, what good is an in silico screen if it cannot be made available at the workbench of drug discovery teams distributed across the globe throughout multi-national pharmaceutical companies? These are not mere discussion points, rather this article embarks on the stepwise mechanics of developing a successful in silico screen. The process is exemplified by our efforts in developing one such screen for predicting metabolic stability of chemicals in a human S9 liver homogenate assay. A real-life use of this in silico screen in a variety of discovery projects at GlaxoSmithKline is presented, highlighting successes and limitations of such applications. Finally, we project some capabilities of in silico ADME tools for greater impact and contribution to successful, efficient drug discovery.

Development and validation of k-nearest-neighbor QSPR models of metabolic stability of drug candidates.

Computational ADME (absorption, distribution, metabolism, and excretion) models may be used early in the drug discovery process in order to flag drug candidates with potentially problematic ADME profiles. We report the development, validation, and application of quantitative structure-property relationship (QSPR) models of metabolic turnover rate for compounds in human S9 homogenate. Biological data were obtained from uniform bioassays of 631 diverse chemicals proprietary to GlaxoSmithKline (GSK). The models were built with topological molecular descriptors such as molecular connectivity indices or atom pairs using the k-nearest neighbor variable selection optimization method developed at the University of North Carolina (Zheng, W.; Tropsha, A. A novel variable selection QSAR approach based on the k-nearest neighbor principle. J. Chem. Inf. Comput. Sci., 2000, 40, 185-194.). For the purpose of validation, the whole data set was divided into training and test sets. The training set QSPR models were characterized by high internal accuracy with leave-one-out cross-validated R(2) (q(2)) values ranging between 0.5 and 0.6. The test set compounds were correctly classified as stable or unstable in S9 assay with an accuracy above 85%. These models were additionally validated by in silico metabolic stability screening of 107 new chemicals under development in several drug discovery programs at GSK. One representative model generated with MolConnZ descriptors predicted 40 compounds to be metabolically stable (turnover rate less than 25%), and 33 of them were indeed found to be stable experimentally. This success (83% concordance) in correctly picking chemicals that are metabolically stable in the human S9 homogenate spells a rapid, computational screen for generating components of the ADME profile in a drug discovery process.

Predicting P-glycoprotein substrates by a quantitative structure-activity relationship model.

A quantitative structure-activity relationship (QSAR) model has been developed to predict whether a given compound is a P-glycoprotein (Pgp) substrate or not. The training set consisted of 95 compounds classified as substrates or non-substrates based on the results from in vitro monolayer efflux assays. The two-group linear discriminant model uses 27 statistically significant, information-rich structure quantifiers to compute the probability of a given structure to be a Pgp substrate. Analysis of the descriptors revealed that the ability to partition into membranes, molecular bulk, and the counts and electrotopological values of certain isolated and bonded hydrides are important structural attributes of substrates. The model fits the data with sensitivity of 100% and specificity of 90.6% in the jackknifed cross-validation test. A prediction accuracy of 86.2% was obtained on a test set of 58 compounds. Examination of the eight "mispredicted" compounds revealed two distinct categories. Five mispredictions were explained by experimental limitations of the efflux assay; these compounds had high permeability and/or were inhibitors of calcein-AM transport. Three mispredictions were due to limitations of the chemical space covered by the current model. The Pgp QSAR model provides an in silico screen to aid in compound selection and in vitro efflux assay prioritization.

Comparison of the computer programs DEREK and TOPKAT to predict bacterial mutagenicity. Deductive Estimate of Risk from Existing Knowledge. Toxicity Prediction by Komputer Assisted Technology.

The performance of two computer programs, DEREK and TOPKAT, was examined with regard to predicting the outcome of the Ames bacterial mutagenicity assay. The results of over 400 Ames tests conducted at Glaxo Wellcome (now GlaxoSmithKline) during the last 15 years on a wide variety of chemical classes were compared with the mutagenicity predictions of both computer programs. DEREK was considered concordant with the Ames assay if (i) the Ames assay was negative (not mutagenic) and no structural alerts for mutagenicity were identified or (ii) the Ames assay was positive (mutagenic) and at least one structural alert was identified. Conversely, the DEREK output was considered discordant if (i) the Ames assay was negative and any structural alert was identified or (ii) the Ames assay was positive and no structural alert was identified. The overall concordance of the DEREK program with the Ames results was 65% and the overall discordance was 35%, based on over 400 compounds. About 23% of the test molecules were outside the permissible limits of the optimum prediction space of TOPKAT. Another 4% of the compounds were either not processable or had indeterminate mutagenicity predictions; these molecules were excluded from the TOPKAT analysis. If the TOPKAT probability was (i) > or =0.7 the molecule was predicted to be mutagenic, (ii) < or =0.3 the compound was predicted to be non-mutagenic and (iii) between 0.3 and 0.7 the prediction was considered indeterminate. From over 300 acceptable predictions, the overall TOPKAT concordance was 73% and the overall discordance was 27%. While the overall concordance of the TOPKAT program was higher than DEREK, TOPKAT fared more poorly than DEREK in the critical Ames-positive category, where 60% of the compounds were incorrectly predicted by TOPKAT as negative but were mutagenic in the Ames test. For DEREK, 54% of the Ames-positive molecules had no structural alerts and were predicted to be non-mutagenic. Alternative methods of analyzing the output of the programs to increase the accuracy with Ames-positive compounds are discussed.