SAR modeling of genotoxic phenomena: the consequence on predictive performance of deviation from a unity ratio of genotoxicants/non-genotoxicants

Food-Borne Chemical Carcinogens and the Evidence for Human Cancer Risk

Commonly consumed foods and beverages can contain chemicals with reported carcinogenic activity in rodent models. Moreover, exposures to some of these substances have been associated with increased cancer risks in humans. Food-borne carcinogens span a range of chemical classes and can arise from natural or anthropogenic sources, as well as form endogenously. Important considerations include the mechanism(s) of action (MoA), their relevance to human biology, and the level of exposure in diet. The MoAs of carcinogens have been classified as either DNA-reactive (genotoxic), involving covalent reaction with nuclear DNA, or epigenetic, involving molecular and cellular effects other than DNA reactivity. Carcinogens are generally present in food at low levels, resulting in low daily intakes, although there are some exceptions. Carcinogens of the DNA-reactive type produce effects at lower dosages than epigenetic carcinogens. Several food-related DNA-reactive carcinogens, including aflatoxins, aristolochic acid, benzene, benzo[a]pyrene and ethylene oxide, are recognized by the International Agency for Research on Cancer (IARC) as causes of human cancer. Of the epigenetic type, the only carcinogen considered to be associated with increased cancer in humans, although not from low-level food exposure, is dioxin (TCDD). Thus, DNA-reactive carcinogens in food represent a much greater risk than epigenetic carcinogens.

In silico prediction of chromosome damage: comparison of three (Q)SAR models

Two major endpoints for genotoxicity tests are gene mutation and chromosome damage (CD), which includes clastogenicity and aneugenicity detected by chromosomal aberration (CA) test or micronucleus (MN) test. Many in silico prediction systems for bacterial mutagenicity (i.e. Ames test results) have been developed and marketed. They show good performance for prediction of Ames mutagenicity. On the other hand, it seems that in silico prediction of CD does not progress as much as Ames prediction. Reasons for this include different mechanisms and detection methods, many false positives and conflicting test results. However, some (quantitative) structure-activity relationship ((Q)SAR) models (e.g. Derek Nexus [Derek], ADMEWorks [AWorks] and CASE Ultra [MCase]) can predict CA test results. Therefore, performances of the three (Q)SAR models were compared using the expanded Carcinogenicity Genotoxicity eXperience (CGX) dataset for understanding current situations and future development. The constructed dataset contained 440 chemicals (325 carcinogens and 115 non-carcinogens). Sensitivity, specificity, accuracy or applicability of each model were 56.0, 86.9, 68.6 or 89.1% in Derek, 67.7, 61.5, 65.2 or 99.3% in AWorks, and 91.0, 64.9, 80.5 or 97.7% in MCase, respectively. The performances (sensitivity and accuracy) of MCase were higher than those of Derek or AWorks. Analysis of predictivity of (Q)SAR models of certain chemical classes revealed no remarkable differences among the models. The tendency of positive prediction by (Q)SAR models was observed in alkylating agents, aromatic amines or amides, aromatic nitro compounds, epoxides, halides and N-nitro or N-nitroso compounds. In an additional investigation, high sensitivity but low specificity was noted in in vivo MN prediction by MCase. Refinement of test data to be used for in silico system (e.g. consideration of cytotoxicity or re-evaluation of conflicting test results) will be needed to improve performance of CD prediction.

CORAL: Building up QSAR models for the chromosome aberration test

A high level of chromosomal aberrations in peripheral blood lymphocytes may be an early marker of cancer risk, but data on risk of specific cancers and types of chromosomal aberrations are limited. Consequently, the development of predictive models for chromosomal aberrations test is important task. Majority of models for chromosomal aberrations test are so-called knowledge-based rules system. The CORAL software (http://www.insilico.eu/coral, abbreviation of “CORrelation And Logic”) is an alternative for knowledge-based rules system. In contrast to knowledge-based rules system, the CORAL software gives possibility to estimate the influence upon the predictive potential of a model of different molecular alerts as well as different splits into the training set and validation set. This possibility is not available for the approaches based on the knowledge-based rules system. Quantitative Structure–Activity Relationships (QSAR) for chromosome aberration test are established for five random splits into the training, calibration, and validation sets. The QSAR approach is based on representation of the molecular structure by simplified molecular input-line entry system (SMILES) without data on physicochemical and/or biochemical parameters. In spite of this limitation, the statistical quality of these models is quite good.

Database composition can affect the structure-activity relationship prediction

The percent active (A) and inactive (I) chemicals in a database can directly affect the sensitivity (% active chemicals predicted correctly) and specificity (% inactive chemicals predicted correctly) of structure-activity relationship (SAR) analyses. Subdividing the National Center for Toxicological Research (NCTR) liver cancer database (NCTRlcdb) into various A/I ratios, which varied from 0.2 to 5.5, resulted in sensitivity/specificity ratios that varied from 0.1 to 6.5. As percent active chemicals increased (increasing A/I ratio), the sensitivity rose, the specificity decreased, and the concordance (% total chemicals predicted correctly) remained fairly constant. The numbers of chemicals in the various data sets ranged from 187 to 999 and appeared to have no affect on any of the 3 predictors of sensitivity, specificity, or concordance.

Classification ensembles for unbalanced class sizes in predictive toxicology

This paper investigates the effects of the ratio of positive-to-negative samples on the sensitivity, specificity, and concordance. When the class sizes in the training samples are not equal, the classification rule derived will favor the majority class and result in a low sensitivity on the minority class prediction. We propose an ensemble classification approach to adjust for differential class sizes in a binary classifier system. An ensemble classifier consists of a set of base classifiers; its prediction rule is based on a summary measure of individual classifications by the base classifiers. Two re-sampling methods, augmentation and abatement, are proposed to generate different bootstrap samples of equal class size to build the base classifiers. The augmentation method balances the two class sizes by bootstrapping additional samples from the minority class, whereas the abatement method balances the two class sizes by sampling only a subset of samples from the majority class. The proposed procedure is applied to a data set to predict estrogen receptor binding activity and to a data set to predict animal liver carcinogenicity using SAR (structure-activity relationship) models as base classifiers. The abatement method appears to perform well in balancing sensitivity and specificity.

Computational prediction of genotoxicity: room for improvement

Decades of mutagenesis and clastogenesis studies have yielded enough structure-activity-relationship (SAR) information to make feasible the construction of computational models for prediction of endpoints based on molecular structure and reactivity. Although there is cause for optimism that these approaches might someday reduce or eliminate the need for actual genotoxicity testing, we are in fact a long way from this. We provide an overview of the state of the art of such approaches, dissecting out how these models are suboptimal. It is clear that current programs still have limited predictive capabilities. We propose that one of the major contributing factors for the inherent lack of sensitivity (typically 50-60%) is inadequate coverage of non-covalent DNA interactions. Suboptimal specificity can be partly attributed to chemical space considerations with associated non-causal activity correlations.