High-Throughput Screening to Advance In Vitro Toxicology: Accomplishments, Challenges, and Future Directions.

Traditionally, chemical toxicity is determined by in vivo animal studies, which are low throughput, expensive, and sometimes fail to predict compound toxicity in humans. Due to the increasing number of chemicals in use and the high rate of drug candidate failure due to toxicity, it is imperative to develop in vitro, high-throughput screening methods to determine toxicity. The Tox21 program, a unique research consortium of federal public health agencies, was established to address and identify toxicity concerns in a high-throughput, concentration-responsive manner using a battery of in vitro assays. In this article, we review the advancements in high-throughput robotic screening methodology and informatics processes to enable the generation of toxicological data, and their impact on the field; further, we discuss the future of assessing environmental toxicity utilizing efficient and scalable methods that better represent the corresponding biological and toxicodynamic processes in humans.

Prediction of chemical-induced acute toxicity using in vitro assay data and chemical structure.

Exposure to various chemicals found in the environment and in the context of drug development can cause acute toxicity. To provide an alternative to in vivo animal toxicity testing, the U.S. Tox21 consortium developed in vitro assays to test a library of approximately 10,000 drugs and environmental chemicals (Tox21 10 K compound library) in a quantitative high-throughput screening (qHTS) approach. In this study, we assessed the utility of Tox21 assay data in comparison with chemical structure information in predicting acute systemic toxicity. Prediction models were developed using four machine learning algorithms, namely Random Forest, Naïve Bayes, eXtreme Gradient Boosting, and Support Vector Machine, and their performance was assessed using the area under the receiver operating characteristic curve (AUC-ROC). The chemical structure-based models as well as the Tox21 assay data demonstrated good predictive power for acute toxicity, achieving AUC-ROC values ranging from 0.83 to 0.93 and 0.73 to 0.79, respectively. We applied the models to predict the acute toxicity potential of the compounds in the Tox21 10 K compound library, most of which were found to be non-toxic. In addition, we identified the Tox21 assays that contributed the most to acute toxicity prediction, such as acetylcholinesterase (AChE) inhibition and p53 induction. Chemical features including organophosphates and carbamates were also identified to be significantly associated with acute toxicity. In conclusion, this study underscores the utility of in vitro assay data in predicting acute toxicity.

In vitro profiling of pesticides within the Tox21 10K compound library for bioactivity and potential toxicity.

Pesticides include a diverse class of toxic chemicals, often having numerous modes of actions when used in agriculture against targeted organisms to control insect infestation, halt unwanted vegetation, and prevent the spread of disease. In this study, the in vitro assay activity of pesticides within the Tox21 10K compound library were examined. The assays in which pesticides showed significantly more activities than non-pesticide chemicals revealed potential targets and mechanisms of action for pesticides. Furthermore, pesticides that showed promiscuous activity against many targets and cytotoxicity were identified, which warrant further toxicological evaluation. Several pesticides were shown to require metabolic activation, demonstrating the importance of introducing metabolic capacity to in vitro assays. Overall, the activity profiles of pesticides highlighted in this study can contribute to the knowledge gaps surrounding pesticide mechanisms and to the better understanding of the on- and off-target organismal effects of pesticides.

Machine Learning-Based Hazard-Driven Prioritization of Features in Nontarget Screening of Environmental High-Resolution Mass Spectrometry Data.

Nontarget high-resolution mass spectrometry screening (NTS HRMS/MS) can detect thousands of organic substances in environmental samples. However, new strategies are needed to focus time-intensive identification efforts on features with the highest potential to cause adverse effects instead of the most abundant ones. To address this challenge, we developed MLinvitroTox, a machine learning framework that uses molecular fingerprints derived from fragmentation spectra (MS2) for a rapid classification of thousands of unidentified HRMS/MS features as toxic/nontoxic based on nearly 400 target-specific and over 100 cytotoxic endpoints from ToxCast/Tox21. Model development results demonstrated that using customized molecular fingerprints and models, over a quarter of toxic endpoints and the majority of the associated mechanistic targets could be accurately predicted with sensitivities exceeding 0.95. Notably, SIRIUS molecular fingerprints and xboost (Extreme Gradient Boosting) models with SMOTE (Synthetic Minority Oversampling Technique) for handling data imbalance were a universally successful and robust modeling configuration. Validation of MLinvitroTox on MassBank spectra showed that toxicity could be predicted from molecular fingerprints derived from MS2 with an average balanced accuracy of 0.75. By applying MLinvitroTox to environmental HRMS/MS data, we confirmed the experimental results obtained with target analysis and narrowed the analytical focus from tens of thousands of detected signals to 783 features linked to potential toxicity, including 109 spectral matches and 30 compounds with confirmed toxic activity.

Identification of nonmonotonic concentration-responses in Tox21 high-throughput screening estrogen receptor assays.

Environmental endocrine-disrupting chemicals (EDCs) interfere with the metabolism and actions of endogenous hormones. It has been well documented in numerous in vivo and in vitro studies that EDCs can exhibit nonmonotonic dose response (NMDR) behaviors. Not conforming to the conventional linear or linear-no-threshold response paradigm, these NMDR relationships pose practical challenges to the risk assessment of EDCs. In the meantime, the endocrine signaling pathways and biological mechanisms underpinning NMDR remain incompletely understood. The US Tox21 program has conducted in vitro cell-based high-throughput screening assays for estrogen receptors (ER), androgen receptors, and other nuclear receptors, and screened the 10 K-compound library for potential endocrine activities. Using 15 concentrations across several orders of magnitude of concentration range and run in both agonist and antagonist modes, these Tox21 assay datasets contain valuable quantitative information that can be explored to evaluate the nonlinear effects of EDCs and may infer potential mechanisms. In this study we analyzed the concentration-response curves (CRCs) in all 8 Tox21 ERα and ERß assays by developing clustering and classification algorithms customized to the datasets to identify various shapes of CRCs. After excluding NMDR curves likely caused by cytotoxicity, luciferase inhibition, or autofluorescence, hundreds of compounds were identified to exhibit Bell or U-shaped CRCs. Bell-shaped CRCs are about 7 times more frequent than U-shaped ones in the Tox21 ER assays. Many compounds exhibit NMDR in at least one assay, and some EDCs well-known for their NMDRs in the literature were also identified, suggesting their nonmonotonic effects may originate at cellular levels involving transcriptional ER signaling. The developed computational methods for NMDR identification in ER assays can be adapted and applied to other high-throughput bioassays.

Prediction of drug-induced liver injury and cardiotoxicity using chemical structure and in vitro assay data.

Drug-induced liver injury (DILI) and cardiotoxicity (DICT) are major adverse effects triggered by many clinically important drugs. To provide an alternative to in vivo toxicity testing, the U.S. Tox21 consortium has screened a collection of ∼10K compounds, including drugs in clinical use, against >70 cell-based assays in a quantitative high-throughput screening (qHTS) format. In this study, we compiled reference compound lists for DILI and DICT and compared the potential of Tox21 assay data with chemical structure information in building prediction models for human in vivo hepatotoxicity and cardiotoxicity. Models were built with four different machine learning algorithms (e.g., Random Forest, Naïve Bayes, eXtreme Gradient Boosting, and Support Vector Machine) and model performance was evaluated by calculating the area under the receiver operating characteristic curve (AUC-ROC). Chemical structure-based models showed reasonable predictive power for DILI (best AUC-ROC = 0.75 ± 0.03) and DICT (best AUC-ROC = 0.83 ± 0.03), while Tox21 assay data alone only showed better than random performance. DILI and DICT prediction models built using a combination of assay data and chemical structure information did not have a positive impact on model performance. The suboptimal predictive performance of the assay data is likely due to insufficient coverage of an adequately predictive number of toxicity mechanisms. The Tox21 consortium is currently expanding coverage of biological response space with additional assays that probe toxicologically important targets and under-represented pathways that may improve the prediction of in vivo toxicity such as DILI and DICT.

Tox21-Based Comparative Analyses for the Identification of Potential Toxic Effects of Environmental Pollutants.

Chemical pollution has become a prominent environmental problem. In recent years, quantitative high-throughput screening (qHTS) assays have been developed for the fast assessment of chemicals' toxic effects. Toxicology in the 21st Century (Tox21) is a well-known and continuously developing qHTS project. Recent reports utilizing Tox21 data have mainly focused on setting up mathematical models for in vivo toxicity predictions, with less attention to intuitive qHTS data visualization. In this study, we attempted to reveal and summarize the toxic effects of environmental pollutants by analyzing and visualizing Tox21 qHTS data. Via PubMed text mining, toxicity/structure clustering, and manual classification, we detected a total of 158 chemicals of environmental concern (COECs) from the Tox21 library that we classified into 13 COEC groups based on structure and activity similarities. By visualizing these COEC groups' bioactivities, we demonstrated that COECs frequently displayed androgen and progesterone antagonistic effects, xenobiotic receptor agonistic roles, and mitochondrial toxicity. We also revealed many other potential targets of the 13 COEC groups, which were not well illustrated yet, and that current Tox21 assays may not correctly classify known teratogens. In conclusion, we provide a feasible method to intuitively understand qHTS data.

Prediction Models for Agonists and Antagonists of Molecular Initiation Events for Toxicity Pathways Using an Improved Deep-Learning-Based Quantitative Structure-Activity Relationship System.

In silico approaches have been studied intensively to assess the toxicological risk of various chemical compounds as alternatives to traditional in vivo animal tests. Among these approaches, quantitative structure-activity relationship (QSAR) analysis has the advantages that it is able to construct models to predict the biological properties of chemicals based on structural information. Previously, we reported a deep learning (DL) algorithm-based QSAR approach called DeepSnap-DL for high-performance prediction modeling of the agonist and antagonist activity of key molecules in molecular initiating events in toxicological pathways using optimized hyperparameters. In the present study, to achieve high throughput in the DeepSnap-DL system-which consists of the preparation of three-dimensional molecular structures of chemical compounds, the generation of snapshot images from the three-dimensional chemical structures, DL, and statistical calculations-we propose an improved DeepSnap-DL approach. Using this improved system, we constructed 59 prediction models for the agonist and antagonist activity of key molecules in the Tox21 10K library. The results indicate that modeling of the agonist and antagonist activity with high prediction performance and high throughput can be achieved by optimizing suitable parameters in the improved DeepSnap-DL system.

Structure-based virtual screening of perfluoroalkyl and polyfluoroalkyl substances (PFASs) as endocrine disruptors of androgen receptor activity using molecular docking and machine learning.

Perfluoroalkyl and polyfluoroalkyl substances (PFASs) pose a substantial threat as endocrine disruptors, and thus early identification of those that may interact with steroid hormone receptors, such as the androgen receptor (AR), is critical. In this study we screened 5,206 PFASs from the CompTox database against the different binding sites on the AR using both molecular docking and machine learning techniques. We developed support vector machine models trained on Tox21 data to classify the active and inactive PFASs for AR using different chemical fingerprints as features. The maximum accuracy was 95.01% and Matthew's correlation coefficient (MCC) was 0.76 respectively, based on MACCS fingerprints (MACCSFP). The combination of docking-based screening and machine learning models identified 29 PFASs that have strong potential for activity against the AR and should be considered priority chemicals for biological toxicity testing.

Use of computational toxicology (CompTox) tools to predict in vivo toxicity for risk assessment.

Computational Toxicology tools were used to predict toxicity for three pesticides: propyzamide (PZ), carbaryl (CB) and chlorpyrifos (CPF). The tools used included: a) ToxCast/Tox21 assays (AC50 s µM: concentration 50% maximum activity); b) in vitro-to-in vivo extrapolation (IVIVE) using ToxCast/Tox21 AC50s to predict administered equivalent doses (AED: mg/kg/d) to compare to known in vivo Lowest-Observed-Effect-Level (LOEL)/Benchmark Dose (BMD); c) high throughput toxicokinetics population based (HTTK-Pop) using AC50s for endpoints associated with the mode of action (MOA) to predict age-adjusted AED for comparison with in vivo LOEL/BMDs. ToxCast/Tox21 active-hit-calls for each chemical were predictive of targets associated with each MOA, however, assays directly relevant to the MOAs for each chemical were limited. IVIVE AEDs were predictive of in vivo LOEL/BMD10s for all three pesticides. HTTK-Pop was predictive of in vivo LOEL/BMD10s for PZ and CPF but not for CB after human age adjustments 11-15 (PZ) and 6-10 (CB) or 6-10 and 11-20 (CPF) corresponding to treated rat ages (in vivo endpoints). The predictions of computational tools are useful for risk assessment to identify targets in chemical MOAs and to support in vivo endpoints. Data can also aid is decisions about the need for further studies.

Molecular Image-Based Prediction Models of Nuclear Receptor Agonists and Antagonists Using the DeepSnap-Deep Learning Approach with the Tox21 10K Library.

The interaction of nuclear receptors (NRs) with chemical compounds can cause dysregulation of endocrine signaling pathways, leading to adverse health outcomes due to the disruption of natural hormones. Thus, identifying possible ligands of NRs is a crucial task for understanding the adverse outcome pathway (AOP) for human toxicity as well as the development of novel drugs. However, the experimental assessment of novel ligands remains expensive and time-consuming. Therefore, an in silico approach with a wide range of applications instead of experimental examination is highly desirable. The recently developed novel molecular image-based deep learning (DL) method, DeepSnap-DL, can produce multiple snapshots from three-dimensional (3D) chemical structures and has achieved high performance in the prediction of chemicals for toxicological evaluation. In this study, we used DeepSnap-DL to construct prediction models of 35 agonist and antagonist allosteric modulators of NRs for chemicals derived from the Tox21 10K library. We demonstrate the high performance of DeepSnap-DL in constructing prediction models. These findings may aid in interpreting the key molecular events of toxicity and support the development of new fields of machine learning to identify environmental chemicals with the potential to interact with NR signaling pathways.

Review of high-content screening applications in toxicology.

High-content screening (HCS) technology combining automated microscopy and quantitative image analysis can address biological questions in academia and the pharmaceutical industry. Various HCS experimental applications have been utilized in the research field of in vitro toxicology. In this review, we describe several HCS application approaches used for studying the mechanism of compound toxicity, highlight some challenges faced in the toxicological community, and discuss the future directions of HCS in regards to new models, new reagents, data management, and informatics. Many specialized areas of toxicology including developmental toxicity, genotoxicity, developmental neurotoxicity/neurotoxicity, hepatotoxicity, cardiotoxicity, and nephrotoxicity will be examined. In addition, several newly developed cellular assay models including induced pluripotent stem cells (iPSCs), three-dimensional (3D) cell models, and tissues-on-a-chip will be discussed. New genome-editing technologies (e.g., CRISPR/Cas9), data analyzing tools for imaging, and coupling with high-content assays will be reviewed. Finally, the applications of machine learning to image processing will be explored. These new HCS approaches offer a huge step forward in dissecting biological processes, developing drugs, and making toxicology studies easier.

In Silico Molecular Docking and In Vivo Validation with Caenorhabditis elegans to Discover Molecular Initiating Events in Adverse Outcome Pathway Framework: Case Study on Endocrine-Disrupting Chemicals with Estrogen and Androgen Receptors.

Molecular docking is used to analyze structural complexes of a target with its ligand for understanding the chemical and structural basis of target specificity. This method has the potential to be applied for discovering molecular initiating events (MIEs) in the Adverse Outcome Pathway framework. In this study, we aimed to develop in silico⁻in vivo combined approach as a tool for identifying potential MIEs. We used environmental chemicals from Tox21 database to identify potential endocrine-disrupting chemicals (EDCs) through molecular docking simulation, using estrogen receptor (ER), androgen receptor (AR) and their homology models in the nematode Caenorhabditis elegans (NHR-14 and NHR-69, respectively). In vivo validation was conducted on the selected EDCs with C. elegans reproductive toxicity assay using wildtype N2, nhr-14, and nhr-69 loss-of-function mutant strains. The chemicals showed high binding affinity to tested receptors and showed the high in vivo reproductive toxicity, and this was further confirmed using the mutant strains. The present study demonstrates that the binding affinity from the molecular docking potentially correlates with in vivo toxicity. These results prove that our in silico⁻in vivo combined approach has the potential to be applied for identifying MIEs. This study also suggests the potential of C. elegans as useful in the in vivo model for validating the in silico approach.

Prediction Model with High-Performance Constitutive Androstane Receptor (CAR) Using DeepSnap-Deep Learning Approach from the Tox21 10K Compound Library.

The constitutive androstane receptor (CAR) plays pivotal roles in drug-induced liver injury through the transcriptional regulation of drug-metabolizing enzymes and transporters. Thus, identifying regulatory factors for CAR activation is important for understanding its mechanisms. Numerous studies conducted previously on CAR activation and its toxicity focused on in vivo or in vitro analyses, which are expensive, time consuming, and require many animals. We developed a computational model that predicts agonists for the CAR using the Toxicology in the 21st Century 10k library. Additionally, we evaluate the prediction performance of novel deep learning (DL)-based quantitative structure-activity relationship analysis called the DeepSnap-DL approach, which is a procedure of generating an omnidirectional snapshot portraying three-dimensional (3D) structures of chemical compounds. The CAR prediction model, which applies a 3D structure generator tool, called CORINA-generated and -optimized chemical structures, in the DeepSnap-DL demonstrated better performance than the existing methods using molecular descriptors. These results indicate that high performance in the prediction model using the DeepSnap-DL approach may be important to prepare suitable 3D chemical structures as input data and to enable the identification of modulators of the CAR.

Toxicity Prediction Method Based on Multi-Channel Convolutional Neural Network.

Molecular toxicity prediction is one of the key studies in drug design. In this paper, a deep learning network based on a two-dimension grid of molecules is proposed to predict toxicity. At first, the van der Waals force and hydrogen bond were calculated according to different descriptors of molecules, and multi-channel grids were generated, which could discover more detail and helpful molecular information for toxicity prediction. The generated grids were fed into a convolutional neural network to obtain the result. A Tox21 dataset was used for the evaluation. This dataset contains more than 12,000 molecules. It can be seen from the experiment that the proposed method performs better compared to other traditional deep learning and machine learning methods.

Improving the Utility of the Tox21 Dataset by Deep Metadata Annotations and Constructing Reusable Benchmarked Chemical Reference Signatures.

The Toxicology in the 21st Century (Tox21) project seeks to develop and test methods for high-throughput examination of the effect certain chemical compounds have on biological systems. Although primary and toxicity assay data were readily available for multiple reporter gene modified cell lines, extensive annotation and curation was required to improve these datasets with respect to how FAIR (Findable, Accessible, Interoperable, and Reusable) they are. In this study, we fully annotated the Tox21 published data with relevant and accepted controlled vocabularies. After removing unreliable data points, we aggregated the results and created three sets of signatures reflecting activity in the reporter gene assays, cytotoxicity, and selective reporter gene activity, respectively. We benchmarked these signatures using the chemical structures of the tested compounds and obtained generally high receiver operating characteristic (ROC) scores, suggesting good quality and utility of these signatures and the underlying data. We analyzed the results to identify promiscuous individual compounds and chemotypes for the three signature categories and interpreted the results to illustrate the utility and re-usability of the datasets. With this study, we aimed to demonstrate the importance of data standards in reporting screening results and high-quality annotations to enable re-use and interpretation of these data. To improve the data with respect to all FAIR criteria, all assay annotations, cleaned and aggregate datasets, and signatures were made available as standardized dataset packages (Aggregated Tox21 bioactivity data, 2019).

Convolutional neural network based on SMILES representation of compounds for detecting chemical motif.

BACKGROUND: Previous studies have suggested deep learning to be a highly effective approach for screening lead compounds for new drugs. Several deep learning models have been developed by addressing the use of various kinds of fingerprints and graph convolution architectures. However, these methods are either advantageous or disadvantageous depending on whether they (1) can distinguish structural differences including chirality of compounds, and (2) can automatically discover effective features. RESULTS: We developed another deep learning model for compound classification. In this method, we constructed a distributed representation of compounds based on the SMILES notation, which linearly represents a compound structure, and applied the SMILES-based representation to a convolutional neural network (CNN). The use of SMILES allows us to process all types of compounds while incorporating a broad range of structure information, and representation learning by CNN automatically acquires a low-dimensional representation of input features. In a benchmark experiment using the TOX 21 dataset, our method outperformed conventional fingerprint methods, and performed comparably against the winning model of the TOX 21 Challenge. Multivariate analysis confirmed that the chemical space consisting of the features learned by SMILES-based representation learning adequately expressed a richer feature space that enabled the accurate discrimination of compounds. Using motif detection with the learned filters, not only important known structures (motifs) such as protein-binding sites but also structures of unknown functional groups were detected. CONCLUSIONS: The source code of our SMILES-based convolutional neural network software in the deep learning framework Chainer is available at http://www.dna.bio.keio.ac.jp/smiles/ , and the dataset used for performance evaluation in this work is available at the same URL.

Infer the in vivo point of departure with ToxCast in vitro assay data using a robust learning approach.

The development and application of high throughput in vitro assays is an important development for risk assessment in the twenty-first century. However, there are still significant challenges to incorporate in vitro assays into routine toxicity testing practices. In this paper, a robust learning approach was developed to infer the in vivo point of departure (POD) with in vitro assay data from ToxCast and Tox21 projects. Assay data from ToxCast and Tox21 projects were utilized to derive the in vitro PODs for several hundred chemicals. These were combined with in vivo PODs from ToxRefDB regarding the rat and mouse liver to build a high-dimensional robust regression model. This approach separates the chemicals into a majority, well-predicted set; and a minority, outlier set. Salient relationships can then be learned from the data. For both mouse and rat liver PODs, over 93% of chemicals have inferred values from in vitro PODs that are within ± 1 of the in vivo PODs on the log10 scale (the target learning region, or TLR) and R2 of 0.80 (rats) and 0.78 (mice) for these chemicals. This is comparable with extrapolation between related species (mouse and rat), which has 93% chemicals within the TLR and the R2 being 0.78. Chemicals in the outlier set tend to also have more biologically variable characteristics. With the continued accumulation of high throughput data for a wide range of chemicals, predictive modeling can provide a valuable complement for adverse outcome pathway based approach in risk assessment.

High-Performance Prediction of Human Estrogen Receptor Agonists Based on Chemical Structures.

Many agonists for the estrogen receptor are known to disrupt endocrine functioning. We have developed a computational model that predicts agonists for the estrogen receptor ligand-binding domain in an assay system. Our model was entered into the Tox21 Data Challenge 2014, a computational toxicology competition organized by the National Center for Advancing Translational Sciences. This competition aims to find high-performance predictive models for various adverse-outcome pathways, including the estrogen receptor. Our predictive model, which is based on the random forest method, delivered the best performance in its competition category. In the current study, the predictive performance of the random forest models was improved by strictly adjusting the hyperparameters to avoid overfitting. The random forest models were optimized from 4000 descriptors simultaneously applied to 10,000 activity assay results for the estrogen receptor ligand-binding domain, which have been measured and compiled by Tox21. Owing to the correlation between our model's and the challenge's results, we consider that our model currently possesses the highest predictive power on agonist activity of the estrogen receptor ligand-binding domain. Furthermore, analysis of the optimized model revealed some important features of the agonists, such as the number of hydroxyl groups in the molecules.

Differential modulation of FXR activity by chlorophacinone and ivermectin analogs.

Chemicals that alter normal function of farnesoid X receptor (FXR) have been shown to affect the homeostasis of bile acids, glucose, and lipids. Several structural classes of environmental chemicals and drugs that modulated FXR transactivation were previously identified by quantitative high-throughput screening (qHTS) of the Tox21 10K chemical collection. In the present study, we validated the FXR antagonist activity of selected structural classes, including avermectin anthelmintics, dihydropyridine calcium channel blockers, 1,3-indandione rodenticides, and pyrethroid pesticides, using in vitro assay and quantitative structural-activity relationship (QSAR) analysis approaches. (Z)-Guggulsterone, chlorophacinone, ivermectin, and their analogs were profiled for their ability to alter CDCA-mediated FXR binding using a panel of 154 coregulator motifs and to induce or inhibit transactivation and coactivator recruitment activities of constitutive androstane receptor (CAR), liver X receptor alpha (LXRα), or pregnane X receptor (PXR). Our results showed that chlorophacinone and ivermectin had distinct modes of action (MOA) in modulating FXR-coregulator interactions and compound selectivity against the four aforementioned functionally-relevant nuclear receptors. These findings collectively provide mechanistic insights regarding compound activities against FXR and possible explanations for in vivo toxicological observations of chlorophacinone, ivermectin, and their analogs.