Leveraging high-throughput screening data,deep neural networks,and conditional generative adversarial networks to advance predictive toxicology

There are currently 85,000 chemicals registered with the Environmental Protection Agency (EPA) under the Toxic Substances Control Act, but only a small fraction have measured toxicological data. To address this gap, high-throughput screening (HTS) and computational methods are vital. As part of one such HTS effort, embryonic zebrafish were used to examine a suite of morphological and mortality endpoints at six concentrations from over 1,000 unique chemicals found in the ToxCast library (phase 1 and 2). We hypothesized that by using a conditional generative adversarial network (cGAN) or deep neural networks (DNN), and leveraging this large set of toxicity data we could efficiently predict toxic outcomes of untested chemicals. Utilizing a novel method in this space, we converted the 3D structural information into a weighted set of points while retaining all information about the structure. In vivo toxicity and chemical data were used to train two neural network generators. The first was a DNN (Go-ZT) while the second utilized cGAN architecture (GAN-ZT) to train generators to produce toxicity data. Our results showed that Go-ZT significantly outperformed the cGAN, support vector machine, random forest and multilayer perceptron models in cross-validation, and when tested against an external test dataset. By combining both Go-ZT and GAN-ZT, our consensus model improved the SE, SP, PPV, and Kappa, to 71.4%, 95.9%, 71.4% and 0.673, respectively, resulting in an area under the receiver operating characteristic (AUROC) of 0.837. Considering their potential use as prescreening tools, these models could provide in vivo toxicity predictions and insight into the hundreds of thousands of untested chemicals to prioritize compounds for HT testing.     

Zebrafish: as an integrative model for twenty-first century toxicity testing

The zebrafish embryo is a useful small model for investigating vertebrate development because of its transparency, low cost, transgenic and morpholino capabilities, conservation of cell signaling, and concordance with mammalian developmental phenotypes. From these advantages, the zebrafish embryo has been considered as an alternative model for traditional in vivo developmental toxicity screening. The use of this organism in conjunction with traditional in vivo developmental toxicity testing has the potential to reduce cost and increase throughput of testing the chemical universe, prioritize chemicals for targeted toxicity testing, generate predictive models of developmental toxicants, and elucidate mechanisms and adverse outcome pathways for abnormal development. This review gives an overview of the zebrafish embryo for pre dictive toxicology and 21st century toxicity testing. Developmental eye defects were selected as an example to evaluate data from the U.S. Environmental Protection Agency's ToxCast program comparing responses in zebrafish embryos with those from pregnant rats and rabbits for a subset of 24 environmental chemicals across >600 in vitro assay targets. Cross-species comparisons implied a common basis for biological pathways associated with neuronal defects, extracellular matrix remodeling, and mitotic arrest.     

Environmental chemical mediated male reproductive toxicity: Drosophila melanogaster as an alternate animal model

Industrialization and indiscriminate use of agrochemicals have increased the human health risk. Recent epidemiological studies raised a concern for male reproduction given their observations of reduced sperm counts and altered semen quality. Interestingly, environmental factors that include various metals, pesticides and their metabolites have been causally linked to such adversities by their presence in the semen at levels that correlate to infertility. The epidemiological observations were further supported by studies in animal models involving various chemicals. Therefore, in this review, we focused on male reproductive toxicity and the adverse effects of different environmental chemicals on male reproduction. However, it is beyond the scope of this review to provide a detailed appraisal of all of the environmental chemicals that have been associated with reproductive toxicity in animals. Here, we provided the evidence for reproductive adversities of some commonly encountered chemicals (pesticides/metals) in the environment. In view of the recent thrust for an alternate to animal models in research, we subsequently discussed the contributions of Drosophila melanogaster as an alternate animal model for quick screening of toxicants for their reproductive toxicity potential. Finally, we emphasized the genetic and molecular tools offered by Drosophila for understanding the mechanisms underlying the male reproductive toxicity.     

Identification of developmental toxicants using the Frog Embryo Teratogenesis Assay-Xenopus (FETAX)

Because growth and development are processes sensitive to the action of many chemicals, bioassays that screen for developmental toxicants may be more indicative of chronic effects than acute toxicity assays. FETAX is a 96 h whole embryo static renewal test employing the embryos of the frog Xenopus laevis. Endpoints are mortality, malformation and growth. Because of the frog's fecundity, its extensive use in basic research and the ability to obtain embryos year-round, it is an ideal organism to use in screening for developmental toxicants. By validating using known mammalian teratogens and the use of rat liver microsomes to stimulate mammalian metabolism, we have extended the use of the system for the prescreening of human developmental toxicants. In past validation work, we have correctly identified the teratogenicity of 15 to 17 compounds used in validation for a predictive accuracy of approximately 88%. In the present study, the ability of FETAX to detect developmental toxicants in groundwater samples taken from an industrial waste dump was evaluated. FETAX showed that it was sensitive enough to detect developmental toxicants in samples without prior concentration. In some samples, less than half the LC50 concentration was required to cause significant malformation. In some cases, a dose-response curve was not obtainable but the test results nonetheless indicated some developmental toxicity. The results of this study indicate that it is necessary to routinely screen for developmental toxicants when establishing water quality criteria for the preservation of species and for human health.     

Using Short-Term Tests to Predict Carcinogenic Activity in the Long-Term Bioassay

A method for classifying chemicals with respect to carcinogenic potential based on short-term test results is presented. The method utilizes the logistic regression model to translate results from short-term toxicity assays into predictions of the likelihood that a chemical will be carcinogenic if tested in a long-term bioassay. The proposed method differs from previous approaches in two ways. First, statistical confidence limits on probabilities of cancer rather than central estimates of those probabilities are used for classification. Second, the method does not classify all chemicals in a data base with respect to carcinogenic potential. Instead, it identifies chemicals with highest and lowest likelihood of testing positive for carcinogenicity in the bioassay. A subset of chemicals with intermediate likelihood of being positive remains unclassified, and will require further testing, perhaps in a long-term bioassay. Two data bases of binary short-term and long-term test results from the literature are used to illustrate and evaluate the proposed procedure. A cross-validation analysis of one of the data sets suggests that, for a sufficiently rich data base of chemicals, the development of a robust predictive system to replace the bioassay for some unknown chemicals is a realistic goal.     

Predictive modeling of whole-cell bioactivity retention data in the presence of organic compounds

The retention of whole-cell bioactivity is essential to the biocatalysis in reaction mixture containing organic compounds for the preparation of products in the pharmaceutical fine chemicals and food industries. A Quantitative Structure-Activity Relationship (QSAR) toxicity model for determination of whole-cell bioactivity was developed on the basis of receptor theory. And the glycolysis activity retention of baker's yeast in the presence of organic compounds was investigated in this study. Based on the fitting of the QSAR model to experimental data obtained in this study and in previous literatures, the regression analysis indicated that calculated values with this model were consistent with experimental data. Comparison of experimental data to the modeled curves suggests good predictive capabilities of the model for determination of whole-cell bioactivity.     

The ECVAM international validation study on in vitro tests for acute skin irritation: report on the validity of the EPISKIN and EpiDerm assays and on the Skin Integrity Function Test

ECVAM sponsored a formal validation study on three in vitro tests for skin irritation, of which two employ reconstituted human epidermis models (EPISKIN, EpiDerm), and one, the skin integrity function test (SIFT), employs ex vivo mouse skin. The goal of the study was to assess whether the in vitro tests would correctly predict in vivo classifications according to the EU classification scheme, "R38" and "no label" (i.e. non-irritant). 58 chemicals (25 irritants and 33 non-irritants) were tested, having been selected to give broad coverage of physico-chemical properties, and an adequate distribution of irritancy scores derived from in vivo rabbit skin irritation tests. In Phase 1, 20 of these chemicals (9 irritants and 11 non-irritants) were tested with coded identities by a single lead laboratory for each of the methods, to confirm the suitability of the protocol improvements introduced after a prevalidation phase. When cell viability (evaluated by the MTT reduction test) was used as the endpoint, the predictive ability of both EpiDerm and EPISKIN was considered sufficient to justify their progression to Phase 2, while the predictive ability of the SIFT was judged to be inadequate. Since both the reconstituted skin models provided false predictions around the in vivo classification border (a rabbit Draize test score of 2), the release of a cytokine, interleukin-1alpha (IL-1alpha), was also determined. In Phase 2, each human skin model was tested in three laboratories, with 58 chemicals. The main endpoint measured for both EpiDerm and EPISKIN was cell viability. In samples from chemicals which gave MTT assay results above the threshold of 50% viability, IL-1alpha release was also measured, to determine whether the additional endpoint would improve the predictive ability of the tests. For EPISKIN, the sensitivity was 75% and the specificity was 81% (MTT assay only); with the combination of the MTT and IL-1alpha assays, the sensitivity increased to 91%, with a specificity of 79%. For EpiDerm, the sensitivity was 57% and the specificity was 85% (MTT assay only), while the predictive capacity of EpiDerm was not improved by the measurement of IL-1alpha release. Following independent peer review, in April 2007 the ECVAM Scientific Advisory Committee endorsed the scientific validity of the EPISKIN test as a replacement for the rabbit skin irritation method, and of the EpiDerm method for identifying skin irritants as part of a tiered testing strategy. This new alternative approach will probably be the first use of in vitro toxicity testing to replace the Draize rabbit skin irritation test in Europe and internationally, since, in the very near future, new EU and OECD Test Guidelines will be proposed for regulatory acceptance.     

Background and overview of current sediment toxicity identification evaluation procedures

Laboratory bioassays can provide an integrated assessment of the potential toxicity of contaminated sediments to aquatic organisms; however, toxicity as a sole endpoint is not particularly useful in terms of identifying remedial options. To focus possible remediation (e.g., source control), it is essential to know which contaminants are responsible for toxicity. Unfortunately, contaminated sediments can contain literally thousands of potentially toxic compounds. Methods which rely solely on correlation to identify contaminants responsible for toxicity are limited in several aspects: (a) actual compounds causing toxicity might not be measured, (b) concentrations of potentially toxic compounds may covary, (c) it may be difficult to assess the bioavailability of contaminants measured in a sediment, and (d) interactions may not be accounted for among potential toxicants (e.g., additivity). Toxicity identification evaluation (TIE) procedures attempt to circumvent these problems by using toxicity-based fractionation procedures to implicate specific contaminants as causative toxicants. Phase I of TIE characterizes the general physio-chemical nature of sample toxicants. Phase II employs methods to measure toxicants via different analytical methods, and Phase III consists of techniques to confirm that the suspect toxicants identified in Phases I and II of the TIE actually are responsible for toxicity. These TIE procedures have been used to investigate the toxicity of a variety of samples, including sediments. Herein we present a brief conceptual overview of the TIE process, and discuss specific considerations associated with sediment TIE research. Points addressed include: (a) selection and preparation of appropriate test fractions, (b) use of benthic organisms for sediment TIE work, and (c) methods for the identification of common sediment contaminants.     

Docking study: PPARs interaction with the selected alternative plasticizers to di(2-ethylhexyl) phthalate

Phthalates, used in medical devices (MDs), have been identified as reproductive and developmental toxicants. Their toxicity varies somewhat depending on the specific phthalate and is in part linked to the activation of Peroxisome Proliferating-Activated Receptors (PPARs). So, the use of MDs containing targeted phthalates such as di(2-ethylhexyl) phthalate (DEHP) has been challenged by European directive 2007/47/EC. Therefore, MDs manufacturers were forced to quickly find replacement plasticizers. However, very little toxicological and epidemiological studies are available on human health. So, we proceeded to dock these chemicals in order to identify compounds that are likely to interact with PPARs binding sites. The results obtained are generally very mixed on the harmlessness of these alternatives. Moreover, no data exist on the biological effects of their possible metabolites. As DEHP toxicity resulted mainly from its major metabolites, generalizing the use of these plasticizers without conducting extensive studies on the possible effects on human health of their metabolites seems inconceivable.     

High-throughput screening assays for the identification of chemical probes

High-throughput screening (HTS) assays enable the testing of large numbers of chemical substances for activity in diverse areas of biology. The biological responses measured in HTS assays span isolated biochemical systems containing purified receptors or enzymes to signal transduction pathways and complex networks functioning in cellular environments. This Review addresses factors that need to be considered when implementing assays for HTS and is aimed particularly at investigators new to this field. We discuss assay design strategies, the major detection technologies and examples of HTS assays for common target classes, cellular pathways and simple cellular phenotypes. We conclude with special considerations for configuring sensitive, robust, informative and economically feasible HTS assays.     

Comprehensive Assessment of Germline Chemical Toxicity Using the Nematode Caenorhabditis elegans

Identifying the reproductive toxicity of the thousands of chemicals present in our environment has been one of the most tantalizing challenges in the field of environmental health. This is due in part to the paucity of model systems that can (1) accurately recapitulate keys features of reproductive processes and (2) do so in a medium- to high-throughput fashion, without the need for a high number of vertebrate animals.We describe here an assay in the nematode C. elegans that allows the rapid identification of germline toxicants by monitoring the induction of aneuploid embryos. By making use of a GFP reporter line, errors in chromosome segregation resulting from germline disruption are easily visualized and quantified by automated fluorescence microscopy. Thus the screening of a particular set of compounds for its toxicity can be performed in a 96- to 384-well plate format in a matter of days. Secondary analysis of positive hits can be performed to determine whether the chromosome abnormalities originated from meiotic disruption or from early embryonic chromosome segregation errors. Altogether, this assay represents a fast first-pass strategy for the rapid assessment of germline dysfunction following chemical exposure.     

A Computational Model to Predict Rat Ovarian Steroid Secretion from In Vitro Experiments with Endocrine Disruptors

A finely tuned balance between estrogens and androgens controls reproductive functions, and the last step of steroidogenesis plays a key role in maintaining that balance. Environmental toxicants are a serious health concern, and numerous studies have been devoted to studying the effects of endocrine disrupting chemicals (EDCs). The effects of EDCs on steroidogenic enzymes may influence steroid secretion and thus lead to reproductive toxicity. To predict hormonal balance disruption on the basis of data on aromatase activity and mRNA level modulation obtained in vitro on granulosa cells, we developed a mathematical model for the last gonadal steps of the sex steroid synthesis pathway. The model can simulate the ovarian synthesis and secretion of estrone, estradiol, androstenedione, and testosterone, and their response to endocrine disruption. The model is able to predict ovarian sex steroid concentrations under normal estrous cycle in female rat, and ovarian estradiol concentrations in adult female rats exposed to atrazine, bisphenol A, metabolites of methoxychlor or vinclozolin, and letrozole.     

Endocrine regulation of the reproduction in crustaceans: Identification of potential targets for toxicants and environmental contaminants

Progress in ecotoxicological research documents that crustaceans are highly vulnerable to diverse chemicals and toxicants in the environment. In particular, pollutants affecting endocrine homeostasis in crustaceans (i.e., endocrine disruptors) are intensively studied, and serious reproductive disorders have been documented. In this review, current knowledge about the endocrine regulation of the crustacean reproduction is put together with the published ecotoxicological data with an attempt to summarize the potential of xenobiotics to affect crustacean reproduction. Following gaps and trends were identified: (1) Studies are required in the field of neurohormone (serotonin and dopamine) regulation of the reproduction and possible modulations by environmental toxicants such as antidepressant drugs. (2) Molting-related parameters (regulated by ecdysteroid hormones) are closely coordinated with the development and reproduction cycles in crustaceans (cross-links with methyl farnesoate signalling), and their susceptibility to toxicants should be studied. (3) Other biochemical targets for xenobiotics were recently discovered in crustaceans and these should be explored by further ecotoxicological studies (e.g., new information about ecdysteroid receptor molecular biology). (4) Some sex steroid hormones known from vertebrates (testosterone, progesterone) have been reported in crustaceans but knowledge about their targets (crustacean steroid receptors) and signalling is still limited. (5) Determination of the sex in developing juveniles (affecting the sex ratio in population) is a sensitive parameter to various xenobiotics (including endocrine disruptors) but its modulation by general environmental stress and non-specific toxicity should be further studied.     

Bacterial motility as a potential tool for rapid toxicity assays

Summary Bacterial motility was evaluated as a potential tool for the rapid assay of the toxicity of chemicals to microorganisms. The level of toxicity was evaluated by the chemical concentration which caused a 50% reduction in motile bacteria at the 60 s of exposure to toxicants. Assay results showed a good correlation with those obtained from a conventional growth inhibition test.     

Computer-aided rodent carcinogenicity prediction

The potential of the computer program PASS (Prediction Activity Spectra for Substances) to predict rodent carcinogenicity for chemical compounds was studied. PASS predicts carcinogenicity of chemical compounds on the basis of their structural formula and of structure-activity relationship analysis of known carcinogens and non-carcinogens. The data on structures and experimental results of 2-year carcinogenicity assays for 412 chemicals from the NTP (National Toxicological Program) and 1190 chemicals from the CPDB (Carcinogenic Potency Database) were used in our study. The predictions take into consideration information about species and sex of animals. For evaluation of the predictive accuracy we used two procedures: leave-one-out cross-validation (LOO CV) and leave-20%-out cross-validation. In the last case we randomly divided the studied data set 20 times into two subsets. The data from the first subset, containing 80% of the compounds, were added to the PASS training set (which includes about 46,000 compounds with about 1500 biological activity types collected during the last 20 years to predict biological activity spectra), the second subset with 20% of the compounds was used as an evaluation set. The mean accuracy of prediction calculated by LOO CV is about 73% for NTP compounds in the 'equivocal' category of carcinogenic activity and 80% for NTP compounds in the 'evidence' category of carcinogenicity. The mean accuracy of prediction for the CPDB database is 89.9% calculated by LOO CV and 63.4% calculated by leave-20%-out cross-validation. Influence of incorporation of species and sex data on the accuracy of carcinogenicity prediction was also investigated. It was shown that the accuracy was increased only for data on male animals.