A comprehensive model for reproductive and developmental toxicity hazard identification: I. Development of a weight of evidence QSAR database

A weight of evidence (WOE) reproductive and developmental toxicology (reprotox) database was constructed that is suitable for quantitative structure-activity relationship (QSAR) modeling and human hazard identification of untested chemicals. The database was derived from multiple publicly available reprotox databases and consists of more than 10,000 individual rat, mouse, or rabbit reprotox tests linked to 2134 different organic chemical structures. The reprotox data were classified into seven general classes (male reproductive toxicity, female reproductive toxicity, fetal dysmorphogenesis, functional toxicity, mortality, growth, and newborn behavioral toxicity), and 90 specific categories as defined in the source reprotox databases. Each specific category contained over 500 chemicals, but the percentage of active chemicals was low, generally only 0.1-10%. The mathematical WOE model placed greater significance on confirmatory observations from repeat experiments, chemicals with multiple findings within a category, and the categorical relatedness of the findings. Using the weighted activity scores, statistical analyses were performed for specific data sets to identify clusters of categories that were correlated, containing similar profiles of active and inactive chemicals. The analysis revealed clusters of specific categories that contained chemicals that were active in two or more mammalian species (trans-species). Such chemicals are considered to have the highest potential risk to humans. In contrast, some specific categories exhibited only single species-specific activities. Results also showed that the rat and mouse were more susceptible to dysmorphogenesis than rabbits (6.1- and 3.6-fold, respectively).     

An analysis of genetic toxicity, reproductive and developmental toxicity, and carcinogenicity data: I. Identification of carcinogens using surrogate endpoints

A retrospective analysis of standard genetic toxicity (genetox) tests, reproductive and developmental toxicity (reprotox) studies, and rodent carcinogenicity bioassays (rcbioassay) was performed to identify the genetox and reprotox endpoints whose results best correlate with rcbioassay observations. A database of 7205 chemicals with genetox (n = 4961), reprotox (n = 2173), and rcbioassay (n = 1442) toxicity data was constructed; 1112 of the chemicals have both genetox and rcbioassay data and 721 chemicals have both reprotox and rcbioassay data. This study differed from previous studies by using conservative weight of evidence criteria to classify chemical carcinogens, data from 63 genetox and reprotox toxicological endpoints, and a new statistical parameter of correlation indicator (CI, the average of specificity and positive predictivity) to identify good surrogate endpoints for predicting carcinogenicity. Among 63 endpoints, results revealed that carcinogenicity was well correlated with certain tests for gene mutation (n = 8), in vivo clastogenicity (n = 2), unscheduled DNA synthesis assay (n = 1), and reprotox (n = 3). The current FDA regulatory battery of four genetox tests used to predict carcinogenicity includes two tests with good correlation (gene mutation in Salmonella and in vivo micronucleus) and two tests with poor correlation (mouse lymphoma gene mutation and in vitro chromosome aberrations) by our criteria.     

An analysis of genetic toxicity, reproductive and developmental toxicity, and carcinogenicity data: II. Identification of genotoxicants, reprotoxicants, and carcinogens using in silico methods

This study examined a novel method to identify carcinogens that employed expanded data sets composed of in silico data pooled with actual experimental genetic toxicity (genetox) and reproductive and developmental toxicity (reprotox) data. We constructed 21 modules using the MC4PC program including 13 of 14 (11 genetox and 3 reprotox) tests that we found correlated with results of rodent carcinogenicity bioassays (rcbioassays) [Matthews, E.J., Kruhlak, N.L., Cimino, M.C., Benz, R.D., Contrera, J.F., 2005b. An analysis of genetic toxicity, reproductive and developmental toxicity, and carcinogenicity data: I. Identification of carcinogens using surrogate endpoints. Regul. Toxicol. Pharmacol.]. Each of the 21 modules was evaluated by cross-validation experiments and those with high specificity (SP) and positive predictivity (PPV) were used to predict activities of the 1442 chemicals tested for carcinogenicity for which actual genetox or reprotox data were missing. The expanded data sets had approximately 70% in silico data pooled with approximately 30% experimental data. Based upon SP and PPV, the expanded data sets showed good correlation with carcinogenicity testing results and had correlation indicator (CI, the average of SP and PPV) values of 75.5-88.7%. Conversely, expanded data sets for 9 non-correlated test endpoints were shown not to correlate with carcinogenicity results (CI values <75%). Results also showed that when Salmonella mutagenic carcinogens were removed from the 12 correlated, expanded data sets, only 7 endpoints showed added value by detecting significantly more additional carcinogens than non-carcinogens.     

Development of quantitative structure-activity relationship (QSAR) models to predict the carcinogenic potency of chemicals. II. Using oral slope factor as a measure of carcinogenic potency

The overall risk associated with exposure to a chemical is determined by combining quantitative estimates of exposure to the chemical with their known health effects. For chemicals that cause carcinogenicity, oral slope factors (OSFs) and inhalation unit risks are used to quantitatively estimate the carcinogenic potency or the risk associated with exposure to the chemical by oral or inhalation route, respectively. Frequently, there is a lack of animal or human studies in the literature to determine OSFs. This study aims to circumvent this problem by developing quantitative structure-activity relationship (QSAR) models to predict the OSFs of chemicals. The OSFs of 70 chemicals based on male/female human, rat, and mouse bioassay data were obtained from the United States Environmental Protection Agency's Integrated Risk Information System (IRIS) database. A global QSAR model that considered all 70 chemicals as well as species and/or sex-specific QSARs were developed in this study. Study results indicate that the species and sex-specific QSARs (r(2)>0.8, q(2)>0.7) had a better predictive abilities than the global QSAR developed using data from all species and sexes (r(2)=0.77, q(2)=0.73). The QSARs developed in this study were externally validated, and demonstrated reasonable predictive abilities.     

Use of the index of ideality of correlation to improve predictive potential for biochemical endpoints

The CORAL software is a tool to build up quantitative structure-property/activity relationships (QSPRs/QSARs). The project of updated version of the CORAL software is discussed in terms of practical applications for building up various models. The updating is the possibility to improve the predictive potential of models using the so-called Index of Ideality of Correlation (IIC) as a criterion of the predictive potential for QSPR/QSAR models. Efficacy of the IIC is examined with three examples of building up QSARs: (i) models for anticancer activity; (ii) models for mutagenicity; and (iii) models for toxicity of psychotropic drugs. The validation of these models has been carried out with several splits into the training, invisible training, calibration, and validation sets. The ability of IIC to be an indicator of predictive potential of QSAR models is confirmed. The updated version of the CORAL software (CORALSEA-2017, http://www.insilico.eu/coral) is available on the Internet.     

In silico approaches to explore toxicity end points: issues and concerns for estimating human health effects

The European Chemicals Bureau and the Organisation for Economic Cooperation and Development are currently compiling a sanctioned list of quantitative structure-activity relationship (QSAR) risk assessment models and data sets to predict the physiological properties, environmental fate, ecological effects and human health effects of new and existing chemicals in commerce in the European Union. This action implements the technical requirements of the European Commission's Registration, Evaluation and Authorisation of Chemicals legislation. The goal is to identify a battery of QSARs that can furnish rapid, reliable and cost-effective decision support information for regulatory decisions that can substitute for results from animal studies. This report discusses issues and concerns that need to be addressed when selecting QSARs to predict human health effect end points.     

Modelling acute oral mammalian toxicity. 1. Definition of a quantifiable baseline effect

Quantitative structure–activity relationships (QSARs) provide a useful tool to define a relationship between chemical structure and toxicity and allow for the prediction of the toxicity of untested chemicals. QSAR models based upon an anaesthetic or narcosis mechanism represent a baseline, or minimum, toxicity, i.e. unless a chemical acts by another, more specific, mechanism, its toxicity will be predicted by such models. The aim of this investigation was to develop baseline models for the acute toxicity of chemicals to mammals (rat and mouse) following the oral route of administration. The availability of such baseline toxicity models for mammalian species can provide a probe for testing new chemicals with respect to their molecular mechanism of toxicity. Multiple-regression type structure–toxicity relationships were derived . (i.e., from oral log data for mammalian species (rat and mouse) and the 1-octanol/water partition coefficient (log P) of classic non-polar narcotics). Subsequently, these models were used to distinguish between reactive chemicals of different mechanistic domains and baseline toxic chemicals. Comparison of measured toxicity data for oral rat and mouse LD₅₀ with predictions from baseline QSAR provides a means of identifying mechanistic categories and for categorising more specific acute mechanisms.     

Prediction of drug-related cardiac adverse effects in humans-B: Use of QSAR programs for early detection of drug-induced cardiac toxicities

This report describes the use of three quantitative structure–activity relationship (QSAR) programs to predict drug-related cardiac adverse effects (AEs), BioEpisteme?, MC4PC, and Leadscope Predictive Data Miner. QSAR models were constructed for 9 cardiac AE clusters affecting Purkinje nerve fibers (arrhythmia, bradycardia, conduction disorder, electrocardiogram, palpitations, QT prolongation, rate rhythm composite, tachycardia, and Torsades de pointes) and 5 clusters affecting the heart muscle (coronary artery disorders, heart failure, myocardial disorders, myocardial infarction, and valve disorders). The models were based on a database of post-marketing AEs linked to 1632 chemical structures, and identical training data sets were configured for three QSAR programs. Model performance was optimized and shown to be affected by the ratio of the number of active to inactive drugs. Results revealed that the three programs were complementary and predictive performances using any single positive, consensus two positives, or consensus three positives were as follows, respectively: 70.7%, 91.7%, and 98.0% specificity; 74.7%, 47.2%, and 21.0% sensitivity; and 138.2, 206.3, and 144.2 χ². In addition, a prospective study using AE data from the U.S. Food and Drug Administration’s (FDA’s) MedWatch Program showed 82.4% specificity and 94.3% sensitivity. Furthermore, an external validation study of 18 drugs with serious cardiotoxicity not considered in the models had 88.9% sensitivity.     

Prediction of rodent carcinogenic potential of naturally occurring chemicals in the human diet using high-throughput QSAR predictive modeling

Consistent with the U.S. Food and Drug Administration (FDA) Critical Path Initiative, predictive toxicology software programs employing quantitative structure-activity relationship (QSAR) models are currently under evaluation for regulatory risk assessment and scientific decision support for highly sensitive endpoints such as carcinogenicity, mutagenicity and reproductive toxicity. At the FDA's Center for Food Safety and Applied Nutrition's Office of Food Additive Safety and the Center for Drug Evaluation and Research's Informatics and Computational Safety Analysis Staff (ICSAS), the use of computational SAR tools for both qualitative and quantitative risk assessment applications are being developed and evaluated. One tool of current interest is MDL-QSAR predictive discriminant analysis modeling of rodent carcinogenicity, which has been previously evaluated for pharmaceutical applications by the FDA ICSAS. The study described in this paper aims to evaluate the utility of this software to estimate the carcinogenic potential of small, organic, naturally occurring chemicals found in the human diet. In addition, a group of 19 known synthetic dietary constituents that were positive in rodent carcinogenicity studies served as a control group. In the test group of naturally occurring chemicals, 101 were found to be suitable for predictive modeling using this software's discriminant analysis modeling approach. Predictions performed on these compounds were compared to published experimental evidence of each compound's carcinogenic potential. Experimental evidence included relevant toxicological studies such as rodent cancer bioassays, rodent anti-carcinogenicity studies, genotoxic studies, and the presence of chemical structural alerts. Statistical indices of predictive performance were calculated to assess the utility of the predictive modeling method. Results revealed good predictive performance using this software's rodent carcinogenicity module of over 1200 chemicals, comprised primarily of pharmaceutical, industrial and some natural products developed under an FDA-MDL cooperative research and development agreement (CRADA). The predictive performance for this group of dietary natural products and the control group was 97% sensitivity and 80% concordance. Specificity was marginal at 53%. This study finds that the in silico QSAR analysis employing this software's rodent carcinogenicity database is capable of identifying the rodent carcinogenic potential of naturally occurring organic molecules found in the human diet with a high degree of sensitivity. It is the first study to demonstrate successful QSAR predictive modeling of naturally occurring carcinogens found in the human diet using an external validation test. Further test validation of this software and expansion of the training data set for dietary chemicals will help to support the future use of such QSAR methods for screening and prioritizing the risk of dietary chemicals when actual animal data are inadequate, equivocal, or absent.     

Comparison of MC4PC and MDL-QSAR rodent carcinogenicity predictions and the enhancement of predictive performance by combining QSAR models

This report presents a comparison of the predictive performance of MC4PC and MDL-QSAR software as well as a method for combining the predictions from both programs to increase overall accuracy. The conclusions are based on 10 x 10% leave-many-out internal cross-validation studies using 1540 training set compounds with 2-year rodent carcinogenicity findings. The models were generated using the same weight of evidence scoring method previously developed [Matthews, E.J., Contrera, J.F., 1998. A new highly specific method for predicting the carcinogenic potential of pharmaceuticals in rodents using enhanced MCASE QSAR-ES software. Regul. Toxicol. Pharmacol. 28, 242-264.]. Although MC4PC and MDL-QSAR use different algorithms, their overall predictive performance was remarkably similar. Respectively, the sensitivity of MC4PC and MDL-QSAR was 61 and 63%, specificity was 71 and 75%, and concordance was 66 and 69%. Coverage for both programs was over 95% and receiver operator characteristic (ROC) intercept statistic values were above 2.00. The software programs had complimentary coverage with none of the 1540 compounds being uncovered by both MC4PC and MDL-QSAR. Merging MC4PC and MDL-QSAR predictions improved the overall predictive performance. Consensus sensitivity increased to 67%, specificity to 84%, concordance to 76%, and ROC to 4.31. Consensus rules can be tuned to reflect the priorities of the user, so that greater emphasis may be placed on predictions with high sensitivity/low false negative rates or high specificity/low false positive rates. Sensitivity was optimized to 75% by reclassifying all compounds predicted to be positive in MC4PC or MDL-QSAR as positive, and specificity was optimized to 89% by reclassifying all compounds predicted negative in MC4PC or MDL-QSAR as negative.     

Comparison of MC4PC and MDL-QSAR rodent carcinogenicity predictions and the enhancement of predictive performance by combining QSAR models

This report presents a comparison of the predictive performance of MC4PC and MDL-QSAR software as well as a method for combining the predictions from both programs to increase overall accuracy. The conclusions are based on 10 × 10% leave-many-out internal cross-validation studies using 1540 training set compounds with 2-year rodent carcinogenicity findings. The models were generated using the same weight of evidence scoring method previously developed [Matthews, E.J., Contrera, J.F., 1998. A new highly specific method for predicting the carcinogenic potential of pharmaceuticals in rodents using enhanced MCASE QSAR-ES software. Regul. Toxicol. Pharmacol. 28, 242–264.]. Although MC4PC and MDL-QSAR use different algorithms, their overall predictive performance was remarkably similar. Respectively, the sensitivity of MC4PC and MDL-QSAR was 61 and 63%, specificity was 71 and 75%, and concordance was 66 and 69%. Coverage for both programs was over 95% and receiver operator characteristic (ROC) intercept statistic values were above 2.00. The software programs had complimentary coverage with none of the 1540 compounds being uncovered by both MC4PC and MDL-QSAR. Merging MC4PC and MDL-QSAR predictions improved the overall predictive performance. Consensus sensitivity increased to 67%, specificity to 84%, concordance to 76%, and ROC to 4.31. Consensus rules can be tuned to reflect the priorities of the user, so that greater emphasis may be placed on predictions with high sensitivity/low false negative rates or high specificity/low false positive rates. Sensitivity was optimized to 75% by reclassifying all compounds predicted to be positive in MC4PC or MDL-QSAR as positive, and specificity was optimized to 89% by reclassifying all compounds predicted negative in MC4PC or MDL-QSAR as negative.     

Comparative analysis of predictive models for nongenotoxic hepatocarcinogenicity using both toxicogenomics and quantitative structure-activity relationships

The primary testing strategy to identify nongenotoxic carcinogens largely relies on the 2-year rodent bioassay, which is time-consuming and labor-intensive. There is an increasing effort to develop alternative approaches to prioritize the chemicals for, supplement, or even replace the cancer bioassay. In silico approaches based on quantitative structure-activity relationships (QSAR) are rapid and inexpensive and thus have been investigated for such purposes. A slightly more expensive approach based on short-term animal studies with toxicogenomics (TGx) represents another attractive option for this application. Thus, the primary questions are how much better predictive performance using short-term TGx models can be achieved compared to that of QSAR models, and what length of exposure is sufficient for high quality prediction based on TGx. In this study, we developed predictive models for rodent liver carcinogenicity using gene expression data generated from short-term animal models at different time points and QSAR. The study was focused on the prediction of nongenotoxic carcinogenicity since the genotoxic chemicals can be inexpensively removed from further development using various in vitro assays individually or in combination. We identified 62 chemicals whose hepatocarcinogenic potential was available from the National Center for Toxicological Research liver cancer database (NCTRlcdb). The gene expression profiles of liver tissue obtained from rats treated with these chemicals at different time points (1 day, 3 days, and 5 days) are available from the Gene Expression Omnibus (GEO) database. Both TGx and QSAR models were developed on the basis of the same set of chemicals using the same modeling approach, a nearest-centroid method with a minimum redundancy and maximum relevancy-based feature selection with performance assessed using compound-based 5-fold cross-validation. We found that the TGx models outperformed QSAR in every aspect of modeling. For example, the TGx models' predictive accuracy (0.77, 0.77, and 0.82 for the 1-day, 3-day, and 5-day models, respectively) was much higher for an independent validation set than that of a QSAR model (0.55). Permutation tests confirmed the statistical significance of the model's prediction performance. The study concluded that a short-term 5-day TGx animal model holds the potential to predict nongenotoxic hepatocarcinogenicity.     

QSARs in ecotoxicological risk assessment

The need for more ecotoxicological data encourages the use of QSARs because of the reduction of (animal) testing, time and cost. QSARs may however only be used if they prove to be reliable and accurate. In this paper, four QSARs were attempted to predict toxicity for 170 compounds from a broad chemical class, using them as a black-box. Predictions were obtained for 122 compounds, indicating an important drawback of QSARs, i.e., for 28% of the compounds QSARs cannot be used at all. Ecosar, Topkat, and QSARs for non-polar and polar narcosis generated predictions for 120, 39, 24, and 11 compounds, respectively. Correlations between experimental and predicted effect concentrations were significant for Topkat and the QSAR for polar narcosis, but generally poor for Ecosar and the QSAR for non-polar narcosis. When predicted effect concentrations for fish were allowed to deviate from experimental values by a factor of 5, correct predictions were generated for 77%, 54%, 68%, and 91% of the compounds using Ecosar, Topkat, and the QSARs for non-polar and polar narcosis, respectively. It was impossible to indicate specific chemical classes for which a QSAR should be used or not. The results show that currently available QSARs cannot be used as a black-box.