Search

As safety is one of the most important properties of drugs, chemical toxicology prediction has received increasing attentions in the drug discovery research. Traditionally, researchers rely on in vitro and in vivo experiments to test the toxicity of chemical compounds. However, not only are these experiments time consuming and costly, but experiments that involve animal testing are increasingly subject to ethical concerns. While traditional machine learning (ML) methods have been used in the field with some success, the limited availability of annotated toxicity data is the major hurdle for further improving model performance. Inspired by the success of semi-supervised learning (SSL) algorithms, we propose a Graph Convolution Neural Network (GCN) to predict chemical toxicity and trained the network by the Mean Teacher (MT) SSL algorithm. Using the Tox21 data, our optimal SSL-GCN models for predicting the twelve toxicological endpoints achieve an average ROC-AUC score of 0.757 in the test set, which is a 6% improvement over GCN models trained by supervised learning and conventional ML methods. Our SSL-GCN models also exhibit superior performance when compared to models constructed using the built-in DeepChem ML methods. This study demonstrates that SSL can increase the prediction power of models by learning from unannotated data. The optimal unannotated to annotated data ratio ranges between 1:1 and 4:1. This study demonstrates the success of SSL in chemical toxicity prediction; the same technique is expected to be beneficial to other chemical property prediction tasks by utilizing existing large chemical databases. Our optimal model SSL-GCN is hosted on an online server accessible through: https://app.cbbio.online/ssl-gcn/home.

A major challenge in drug development is safety and toxicity concerns due to drug side effects. One such side effect, drug-induced liver injury (DILI), is considered a primary factor in regulatory clearance. The Critical Assessment of Massive Data Analysis (CAMDA) 2020 CMap Drug Safety Challenge goal was to develop prediction models based on gene perturbation of six preselected cell-lines (CMap L1000), extended structural information (MOLD2), toxicity data (TOX21), and FDA reporting of adverse events (FAERS). Four types of DILI classes were targeted, including two clinically relevant scores and two control classifications, designed by the CAMDA organizers. The L1000 gene expression data had variable drug coverage across cell lines with only 247 out of 617 drugs in the study measured in all six cell types. We addressed this coverage issue by using Kru-Bor ranked merging to generate a singular drug expression signature across all six cell lines. These merged signatures were then narrowed down to the top and bottom 100, 250, 500, or 1,000 genes most perturbed by drug treatment. These signatures were subject to feature selection using Fisher’s exact test to identify genes predictive of DILI status. Models based solely on expression signatures had varying results for clinical DILI subtypes with an accuracy ranging from 0.49 to 0.67 and Matthews Correlation Coefficient (MCC) values ranging from -0.03 to 0.1. Models built using FAERS, MOLD2, and TOX21 also had similar results in predicting clinical DILI scores with accuracy ranging from 0.56 to 0.67 with MCC scores ranging from 0.12 to 0.36. To incorporate these various data types with expression-based models, we utilized soft, hard, and weighted ensemble voting methods using the top three performing models for each DILI classification. These voting models achieved a balanced accuracy up to 0.54 and 0.60 for the clinically relevant DILI subtypes. Overall, from our experiment, traditional machine learning approaches may not be optimal as a classification method for the current data.

To screen the tens of thousands of chemicals for which no toxicity data currently exists, it is necessary to move from in vivo rodent models to alternative models, such as zebrafish. Here, we used dechorionated Tropical 5D wild-type zebrafish embryos to screen a 91-compound library provided by the National Toxicology Program (NTP) for developmental toxicity. This library contained 86 unique chemicals that included negative controls, flame retardants, polycyclic aromatic hydrocarbons (PAHs), drugs, industrial chemicals, and pesticides. Fish were exposed to 5 concentrations of each chemical or an equal amount of vehicle (0.5% DMSO) in embryo medium from 6 h post-fertilization (hpf) to 5 days post-fertilization (dpf). Fish were examined daily for mortality and teratogenic effects and photomotor behavior was assessed at 4 and 5 dpf. Of the 5 negative control compounds in the library, none caused mortality/teratogenesis, but two altered behavior. Chemicals provided in duplicate produced similar outcomes. Overall, 13 compounds caused mortality/teratology but not behavioral abnormalities, 24 only affected behavior, and 18 altered both endpoints, with behavior affected at concentrations that did not cause mortality/teratology (55/86 hits). Of the compounds that affected behavior, 52% caused behavioral abnormalities at either 4 or 5 dpf. Compounds within the same functional group caused different behavioral abnormalities, while similar behavioral patterns were caused by compounds from different groups. Our data suggest that behavior is a sensitive endpoint for developmental toxicity screening that integrates multiple modes of toxic action and is influenced by the age of the larval fish at the time of testing.

Humans are exposed to tens of thousands of chemicals over the course of a lifetime, yet there remains inadequate data on the potential harmful effects of these substances on human health. Using quantitative high-throughput screening (qHTS), we can test these compounds for potential toxicity in a more efficient and cost-effective way compared to traditional animal studies. Tox21 has developed a library of ~10,000 chemicals (Tox21 10K) comprising one-third approved and investigational drugs and two-thirds environmental chemicals. In this study, the Tox21 10K was screened in a qHTS format against a panel of 70 cell-based assays with 213 readouts covering a broad range of biological pathways. Activity profiles were compared with chemical structure to assess their ability to differentiate drugs from environmental chemicals, and structure was found to be a better predictor of which chemicals are likely to be drugs. Drugs and environmental chemicals were further analyzed for diversity in structure and biological response space and showed distinguishable, but not distinct, responses in the Tox21 assays. Inclusion of other targets and pathways to further diversify the biological response space covered by these assays is likely required to better evaluate the safety profile of drugs and environmental chemicals to prioritize for in-depth toxicological studies.

Introduction: Recent nationwide surveys found that natural products, including botanical dietary supplements, are used by ∼18% of adults. In many cases, there is a paucity of toxicological data available for these substances to allow for confident evaluations of product safety. The National Toxicology Program (NTP) has received numerous nominations from the public and federal agencies to study the toxicological effects of botanical dietary supplements. The NTP sought to evaluate the utility of in vitro quantitative high-throughput screening (qHTS) assays for toxicological assessment of botanical and dietary supplements. Materials and Methods: In brief, concentration–response assessments of 90 test substances, including 13 distinct botanical species, and individual purported active constituents were evaluated using a subset of the Tox21 qHTS testing panel. The screen included 20 different endpoints that covered a broad range of biologically relevant signaling pathways to detect test article effects upon endocrine activity, nuclear receptor signaling, stress response signaling, genotoxicity, and cell death signaling. Results and Discussion: Botanical dietary supplement extracts induced measurable and diverse activity. Elevated biological activity profiles were observed following treatments with individual chemical constituents relative to their associated botanical extract. The overall distribution of activity was comparable to activities exhibited by compounds present in the Tox21 10K chemical library. Conclusion: Botanical supplements did not exhibit minimal or idiosyncratic activities that would preclude the use of qHTS platforms as a feasible method to screen this class of compounds. However, there are still many considerations and further development required when attempting to use in vitro qHTS methods to characterize the safety profile of botanical/dietary supplements.

The US Environmental Protection Agency Endocrine Disruptor Screening Program (EDSP) is a tiered screening approach to determine the potential for a chemical to interact with estrogen, androgen, or thyroid hormone systems and/or perturb steroidogenesis. Use of high-throughput screening (HTS) to predict hazard and exposure is shifting the EDSP approach to (1) prioritization of chemicals for further screening; and (2) targeted use of EDSP Tier 1 assays to inform specific data needs. In this work, toxicology data for three triazole fungicides (triadimefon, propiconazole, and myclobutanil) were evaluated, including HTS results, EDSP Tier 1 screening (and other scientifically relevant information), and EPA guideline mammalian toxicology study data. The endocrine-related bioactivity predictions from HTS and information that satisfied the EDSP Tier 1 requirements were qualitatively concordant. Current limitations in the available HTS battery for thyroid and steroidogenesis pathways were mitigated by inclusion of guideline toxicology studies in this analysis. Similar margins (3–5 orders of magnitude) were observed between HTS-predicted human bioactivity and exposure values and between in vivo mammalian bioactivity and EPA chronic human exposure estimates for these products’ registered uses. Combined HTS hazard and human exposure predictions suggest low priority for higher-tiered endocrine testing of these triazoles. Comparison with the mammalian toxicology database indicated that this HTS-based prioritization would have been protective for any potential in vivo effects that form the basis of current risk assessment for these chemicals. This example demonstrates an effective, human health protective roadmap for EDSP evaluation of pesticide active ingredients via prioritization using HTS and guideline toxicology information.

The National Toxicology Program Interagency Center for the Evaluation of Alternative Toxicological Methods (NICEATM) provides validation support for US Federal agencies and the US Tox21 interagency consortium, an interagency collaboration that is using high throughput screening (HTS) and other advanced approaches to better understand and predict chemical hazards to humans and the environment. The use of HTS data from assays relevant to the estrogen receptor signaling data pathway is used as an example of how HTS data can be combined with computational modeling to meet the needs of US agencies. As brief summary of US efforts in the areas of biologics testing, acute toxicity, and skin sensitization will also be provided.

Recent efforts aimed at integrating in vitro high-throughput screening (HTS) data into chemical toxicity assessments are necessitating increased understanding of concordance between chemical-induced responses observed in vitro versus in vivo. This investigation set out to (1) measure concordance between in vitro HTS data and transcriptomic responses observed in vivo, focusing on the liver, and (2) identify attributes that can influence concordance. Signal response profiles from 130 substances were compared between in vitro data produced through Tox21 and liver transcriptomic data through DrugMatrix, collected from rats exposed to a chemical for ≤5 days. A global in vitro-to-in vivo comparative analysis based on pathway-level responses resulted in an overall average percent agreement of 79%, ranging on a per-chemical basis between 41% and 100%. Whereas concordance amongst inactive chemicals was high (89%), concordance amongst chemicals showing in vitro activity was only 13%, suggesting that follow-up in vivo and/or orthogonal in vitro assays would improve interpretations of in vitro activity. Attributes identified to influence concordance included experimental design attributes (eg, cell type), target pathways, and physicochemical properties (eg, logP). The attribute that most consistently increased concordance was dose applicability, evaluated by filtering for experimental doses administered to rats that were within 10-fold of those related to likely bioactivity, derived using Tox21 data and high-throughput toxicokinetic modeling. Together, findings suggest that in vitro screening approaches to predict in vivo toxicity are viable particularly when certain attributes are considered, including whether activity versus inactivity is observed, experimental design, chemical properties, and dose applicability.

Increase in the number of new chemicals synthesized in past decades has resulted in constant growth in the development and application of computational models for prediction of activity as well as safety profiles of the chemicals. Most of the time, such computational models and its application must deal with imbalanced chemical data. It is indeed a challenge to construct a classifier using imbalanced data set. In this study, we analyzed and validated the importance of different sampling methods over non-sampling method, to achieve a well-balanced sensitivity and specificity of a machine learning model trained on imbalanced chemical data. Additionally, this study has achieved an accuracy of 93.00%, an AUC of 0.94, F1 measure of 0.90, sensitivity of 96.00% and specificity of 91.00% using SMOTE sampling and Random Forest classifier for the prediction of Drug Induced Liver Injury (DILI). Our results suggest that, irrespective of data set used, sampling methods can have major influence on reducing the gap between sensitivity and specificity of a model. This study demonstrates the efficacy of different sampling methods for class imbalanced problem using binary chemical data sets.

Highlights • RPTEC/TERT1 cells are sensitive to low levels of cadmium and benzo[a]pyrene. • We demonstrate compound specific gene expression responses after co-exposure. • Increased CYP activity was detected but not altered in the presence of cadmium. • RPTEC/TERT1 cells mirror responses expected in the proximal tubule cells in vivo. • The RPTEC/TERT1 cell line provides a biologically relevant in vitro model., We have characterized initial canonical responses to two environmental toxicants, cadmium (Cd) and benzo[a]pyrene (B[a]P), in a novel in vitro model derived from renal proximal tubule epithelial cells (RPTEC) of a healthy human donor. The RPTEC/TERT1 cell line has been immortalized using the human telomerase reverse transcriptase (hTERT) subunit only and does not exhibit chromosomal abnormalities. RPTEC/TERT1 cells were exposed to single-compound and binary mixtures of Cd and B[a]P, known or suspected renal toxicants respectively. Cells exhibited cytotoxicity to concentrations of B[a]P and Cd as low as 1 nm and 3 μM, respectively. RPTEC/TERT1 cells exhibited compound-specific gene expression responses when exposed to 0.01–1 μM B[a]P and 0.1–10 μM Cd. A significant increase in the expression of genes coding for B[a]P metabolizing enzymes (CYP1A1, CYP1B1) occurred in a dose and time dependent manner at 3, 6, and 24 h post exposure. Likewise, a significant increase in the heavy metal responsive gene MT2A was observed following exposure to Cd. The EROD activity assay confirmed significant increases in CYP1(A/B) activity after 24 h of exposure to B[a]P which was not affected by the presence of Cd. Co-exposure to low concentrations of Cd and B[a]P were consistent with changes in gene expression as seen with single-compound exposures. These experiments are the first to provide information regarding toxicological responses in the RPTEC/TERT1 cell line that model those of the target tissue. We conclude that these cells can provide a useful tool for future toxicological studies.

Background: In 2008, the National Institute of Environmental Health Sciences/National Toxicology Program, the U.S. Environmental Protection Agency’s National Center for Computational Toxicology, and the National Human Genome Research Institute/National Institutes of Health Chemical Genomics Center entered into an agreement on “high throughput screening, toxicity pathway profiling, and biological interpretation of findings.” In 2010, the U.S. Food and Drug Administration (FDA) joined the collaboration, known informally as Tox21. Objectives: The Tox21 partners agreed to develop a vision and devise an implementation strategy to shift the assessment of chemical hazards away from traditional experimental animal toxicology studies to one based on target-specific, mechanism-based, biological observations largely obtained using in vitro assays. Discussion: Here we outline the efforts of the Tox21 partners up to the time the FDA joined the collaboration, describe the approaches taken to develop the science and technologies that are currently being used, assess the current status, and identify problems that could impede further progress as well as suggest approaches to address those problems. Conclusion: Tox21 faces some very difficult issues. However, we are making progress in integrating data from diverse technologies and end points into what is effectively a systems-biology approach to toxicology. This can be accomplished only when comprehensive knowledge is obtained with broad coverage of chemical and biological/toxicological space. The efforts thus far reflect the initial stage of an exceedingly complicated program, one that will likely take decades to fully achieve its goals. However, even at this stage, the information obtained has attracted the attention of the international scientific community, and we believe these efforts foretell the future of toxicology.

The need for filling information gaps while reducing toxicity testing in animals is becoming more predominant in risk assessment. Recent legislations are accepting in silico approaches for predicting toxicological outcomes. This article describes the results of Quantitative Structure Activity Relationship (QSAR) modeling efforts within Tox21 Data Challenge 2014 1 , which calculated the best balanced accuracy across all molecular pathway endpoints as well as the highest scores for ATAD5 and mitochondrial membrane potential disruption. Automated QSPR workflow systems, OCHEM (http://ochem.eu), the analytics platform, KNIME and the statistics software, CRAN R, were used to conduct the analysis and develop consensus models using 10 different descriptor sets. A detailed analysis of QSAR models for all 12 molecular pathways and the effect of underlying models' accuracy on the quality of the consensus model are provided. The resulting consensus models yielded a balanced accuracy as high as 88.1% ± 0.6 for mitochondrial membrane disruptors. Such high balanced accuracy and use of the applicability domain show a promising potential for in silico modeling to complement design HTS screening experiments. The comprehensive statistics of all models are publicly available online at https://github.com/amaziz/Tox21-Challenge-Publication while the developed consensus models can be accessed at http://ochem.eu/article/98009.

The Tox21 Data Challenge has been the largest effort of the scientific community to compare computational methods for toxicity prediction. This challenge comprised 12,000 environmental chemicals and drugs which were measured for 12 different toxic effects by specifically designed assays. We participated in this challenge to assess the performance of Deep Learning in computational toxicity prediction. Deep Learning has already revolutionized image processing, speech recognition, and language understanding but has not yet been applied to computational toxicity. Deep Learning is founded on novel algorithms and architectures for artificial neural networks together with the recent availability of very fast computers and massive datasets. It discovers multiple levels of distributed representations of the input, with higher levels representing more abstract concepts. We hypothesized that the construction of a hierarchy of chemical features gives Deep Learning the edge over other toxicity prediction methods. Furthermore, Deep Learning naturally enables multi-task learning, that is, learning of all toxic effects in one neural network and thereby learning of highly informative chemical features. In order to utilize Deep Learning for toxicity prediction, we have developed the DeepTox pipeline. First, DeepTox normalizes the chemical representations of the compounds. Then it computes a large number of chemical descriptors that are used as input to machine learning methods. In its next step, DeepTox trains models, evaluates them, and combines the best of them to ensembles. Finally, DeepTox predicts the toxicity of new compounds. In the Tox21 Data Challenge, DeepTox had the highest performance of all computational methods winning the grand challenge, the nuclear receptor panel, the stress response panel, and six single assays (teams ``Bioinf@JKU''). We found that Deep Learning excelled in toxicity prediction and outperformed many other computational approaches like naive Bayes, support vector machines, and random forests.

Rapid high throughput in vitro screening (HTS) assays are now available for characterizing dose–responses in assays that have been selected for their sensitivity in detecting estrogen-related endpoints. For example, EPA’s ToxCast™ program recently released endocrine assay results for more than 1800 substances and the interagency Tox21 consortium is in the process of releasing data for approximately 10,000 chemicals. But such activity measurements alone fall short for the purposes of priority setting or screening because the relevant exposure context is not considered. Here, we extend the method of exposure:activity profiling by calculating the exposure:activity ratios (EARs) using human exposure estimates and AC50 values for a range of chemicals tested in a suite of seven estrogenic assays in ToxCast™ and Tox21. To provide additional context, relative estrogenic exposure:activity quotients (REEAQ) were derived by comparing chemical-specific EARs to the EAR of the ubiquitous dietary phytoestrogen, genistein (GEN). Although the activity of a substance in HTS-endocrine assays is not a measure of health hazard or risk, understanding how such a dose compares to human exposures provides a valuable additional metric that can be used in decision-making; substances with small EARs and REEAQs would indicate low priority for further endocrine screening or testing.

Background: Oxidative stress has been implicated in the pathogenesis of a variety of diseases ranging from cancer to neurodegeneration, highlighting the need to identify chemicals that can induce this effect. The antioxidant response element (ARE) signaling pathway plays an important role in the amelioration of oxidative stress. Thus, assays that detect the up-regulation of this pathway could be useful for identifying chemicals that induce oxidative stress. Objectives: We used cell-based reporter methods and informatics tools to efficiently screen a large collection of environmental chemicals and identify compounds that induce oxidative stress. Methods: We utilized two cell-based ARE assay reporters, β-lactamase and luciferase, to screen a U.S. National Toxicology Program 1,408-compound library (NTP 1408, which contains 1,340 unique compounds) for their ability to induce oxidative stress in HepG2 cells using quantitative high throughput screening (qHTS). Results: Roughly 3% (34 of 1,340) of the unique compounds demonstrated activity across both cell-based assays. Based on biological activity and structure–activity relationship profiles, we selected 50 compounds for retesting in the two ARE assays and in an additional follow-up assay that employed a mutated ARE linked to β-lactamase. Using this strategy, we identified 30 compounds that demonstrated activity in the ARE-bla and ARE-luc assays and were able to determine structural features conferring compound activity across assays. Conclusions: Our results support the robustness of using two different cell-based approaches for identifying compounds that induce ARE signaling. Together, these methods are useful for prioritizing chemicals for further in-depth mechanism-based toxicity testing.