The Tox21 10K Compound Library: Collaborative Chemistry Advancing Toxicology

Interpretation of thyroid-relevant bioactivity data for comparison to in vivo exposures: A prioritization approach for putative chemical inhibitors of in vitro deiodinase activity

Many ToxCast assay endpoints can be mapped to molecular initiating events (MIEs) within the thyroid adverse outcome pathway (AOP) network. Herein, we provide a framework for interpretation of thyroid-relevant bioactivity data across MIEs. As a proof-of-concept, we used ToxCast data on the inhibition of deiodinase (DIO) enzymes, which convert thyroid hormones between active and inactive forms, and identified substances most likely to inhibit DIO enzymes. Data from 4 relevant cell-free in vitro assays are available for > 2000 chemicals in single concentration screening and 327 chemicals in multi-concentration screening. We filtered to identify chemicals that demonstrated inhibition for each DIO enzyme less likely to be confounded by assay interference, refining the list of putatively active chemicals from 523 to 135. In vitro bioactivity data were then used to estimate administered equivalent doses (AEDs) using a novel high-throughput toxicokinetic (HTTK) model for in vitro to in vivo extrapolation (IVIVE) of dose. To consider potential thyroid-disrupting activity in an appropriate life-stage and dose context, we extended an existing human maternal-fetal HTTK model to allow for simulations involving the first trimester of pregnancy. For many chemicals, using modeled fetal tissue concentrations produced lower AED estimates than using modeled maternal plasma concentrations alone, at least partially due to conservative assumptions in our HTTK model of complete gestation. This extensible approach for MIE groups of thyroid-related bioactivity data from ToxCast may inform further screening or analyses for potential adverse outcomes during pregnancy and development.

Application of cytochrome P450 enzyme assays to predict p53 inducers and AChE inhibitors that require metabolic activation

Metabolically active compounds can cause toxicity which would otherwise be undetected using traditional in vitro assays with limited proficiency for xenobiotic metabolism. Introduction of liver microsomes to assay systems enables enhanced identification of compounds that require biotransformation to induce toxicity. Previously, metabolically active compounds from the Tox21 10 K compound library were identified using assays probing two targets, p53 and acetylcholinesterase (AChE), in the presence and absence of human or rat liver microsomes, due to the established roles of cytochrome P450 (CYP) enzymes in human drug metabolism. To further explore the role of metabolic activation, the activities of the identified metabolically active compounds were evaluated against five CYP enzymes: CYP1A2, CYP2C9, CYP2C19, CYP2D6, and CYP3A4. CYP bioactivities were found to be highly predictive (>80 % accuracy) of compounds that required metabolic activation in these assays. Chemical features significantly enriched in metabolically active compounds, as well as chemical features that were specific for each of the five CYPs, were identified. Product use exposures of the metabolically active compounds were examined in this study, with "pesticides" appearing to be the largest category that may produce harmful metabolites. Additionally, the compound interactions with different CYPs were assessed and frequencies for both classes of compounds, drugs and environmental chemicals, were found to be proportionally similar across the five CYP isoforms.

Machine Learning for Toxicity Prediction Using Chemical Structures: Pillars for Success in the Real World

Machine learning (ML) is increasingly valuable for predicting molecular properties and toxicity in drug discovery. However, toxicity-related end points have always been challenging to evaluate experimentally with respect to in vivo translation due to the required resources for human and animal studies; this has impacted data availability in the field. ML can augment or even potentially replace traditional experimental processes depending on the project phase and specific goals of the prediction. For instance, models can be used to select promising compounds for on-target effects or to deselect those with undesirable characteristics (e.g., off-target or ineffective due to unfavorable pharmacokinetics). However, reliance on ML is not without risks, due to biases stemming from nonrepresentative training data, incompatible choice of algorithm to represent the underlying data, or poor model building and validation approaches. This might lead to inaccurate predictions, misinterpretation of the confidence in ML predictions, and ultimately suboptimal decision-making. Hence, understanding the predictive validity of ML models is of utmost importance to enable faster drug development timelines while improving the quality of decisions. This perspective emphasizes the need to enhance the understanding and application of machine learning models in drug discovery, focusing on well-defined data sets for toxicity prediction based on small molecule structures. We focus on five crucial pillars for success with ML-driven molecular property and toxicity prediction: (1) data set selection, (2) structural representations, (3) model algorithm, (4) model validation, and (5) translation of predictions to decision-making. Understanding these key pillars will foster collaboration and coordination between ML researchers and toxicologists, which will help to advance drug discovery and development.

Developmental toxicity: artificial intelligence-powered assessments

Regulatory agencies require comprehensive toxicity testing for prenatal drug exposure, including new drugs in development, to reduce concerns about developmental toxicity, that is, drug-induced toxicity and adverse effects in pregnant women and fetuses. However, defining developmental toxicity endpoints and optimal analysis of associated public big data remain challenging. Recently, artificial intelligence (AI) approaches have had a critical role in analyzing complex, high-dimensional data, uncovering subtle relationships between chemical exposures and associated developmental risks. Here, we present an overview of major big data resources and data-driven models that focus on predicting various toxicity endpoints. We also highlight emerging, interpretable AI models that integrate multimodal data and domain knowledge to reveal toxic mechanisms underlying complex endpoints, and outline a potential framework that leverages multiple interpretable models to comprehensively evaluate chemical-induced developmental toxicity.

Artificial intelligence approaches for anti-addiction drug discovery

Drug addiction remains a complex global public health challenge, with traditional anti-addiction drug discovery hindered by limited efficacy and slow progress in targeting intricate neurochemical systems. Advanced algorithms within artificial intelligence (AI) present a transformative solution that boosts both speed and precision in therapeutic development. This review examines how artificial intelligence serves as a crucial element in developing anti-addiction medications by targeting the opioid system along with dopaminergic and GABAergic systems, which are essential in addiction pathology. It identifies upcoming trends promising in studying less-researched addiction-linked systems through innovative general-purpose drug discovery techniques. AI holds the potential to transform anti-addiction research by breaking down conventional limitations, which will enable the development of superior treatment methods.

Novel target identification towards drug repurposing based on biological activity profiles

Rare diseases affect more than 30 million individuals, with the majority facing limited treatment options, elevating the urgency to innovative therapeutic solutions. Addressing these medical challenges necessitates an exploration of novel treatment modalities. Among these, drug repurposing emerges as a promising avenue, offering both potential and risk mitigation. To achieve this goal, we primarily focused on developing predictive models that harness cutting-edge computational techniques to uncover latent relationships between gene targets and chemical compounds towards drug repurposing. Building upon our previous investigation, where we successfully identified gene targets for compounds from the Tox21 in vitro assays, our endeavor expanded to a systematic prediction of potential targets for drug repurposing employing machine learning models built on diverse algorithms such as Support Vector Classifier, K-Nearest Neighbors, Random Forest, and Extreme Gradient Boosting. These models were trained on comprehensive biological activity profile data to predict the relationship between 143 gene targets and over 6000 compounds. Our models demonstrated high accuracy (>0.75), with predictions further validated by using public experimental datasets. Furthermore, several findings were evaluated via case studies. By elucidating these connections, we aim to streamline the drug repurposing process, ultimately catalyzing the discovery of more effective therapeutic interventions for rare diseases.

Cell Painting PLUS: expanding the multiplexing capacity of Cell Painting-based phenotypic profiling using iterative staining-elution cycles

Phenotypic changes in the morphology and internal organization of cells can indicate perturbations in cell functions. Therefore, imaging-based high-throughput phenotypic profiling (HTPP) applications such as Cell Painting (CP) play an important role in basic and translational research, drug discovery, and regulatory toxicology. Here we present the Cell Painting PLUS (CPP) assay, an efficient, robust and broadly applicable approach that further expands the versatility of available HTPP methods and offers additional options for addressing mode-of-action specific research questions. An iterative staining-elution cycle allows multiplexing of at least seven fluorescent dyes that label nine different subcellular compartments and organelles including the plasma membrane, actin cytoskeleton, cytoplasmic RNA, nucleoli, lysosomes, nuclear DNA, endoplasmic reticulum, mitochondria, and Golgi apparatus. In this way, CPP significantly expands the flexibility, customizability, and multiplexing capacity of the original CP method and, importantly, also improves the organelle-specificity and diversity of the phenotypic profiles due to the separate imaging and analysis of single dyes in individual channels.

The Missing Enzymes: A Call to Update Pharmacological Profiling Practices for Better Drug Safety Assessment

Pharmacological profiling is critical for the development of safe drugs. With increasing awareness of its significance and attempts to share best practices, here we aimed to understand how pharmacological profiling is implemented and reported in the primary literature by analyzing the representation of nonkinase enzymes in selectivity screens. This aspect has been overlooked in previous publications, despite enzymes constituting a significant portion of the pharmacological targets for currently marketed drugs. Our analysis shows that while industry recommendations for improved pharmacological profiling have been widely adopted, enzymes remain largely underrepresented: about a quarter of studies did not include enzymes, and on average, enzymes comprise only 11% of all targets in pharmacological screens. We discuss possible reasons for this shortcoming and provide examples of critical enzymes missing from current screens. We conclude with the notion that selectivity screens should be expanded to include more enzymes to improve drug development and safety.

Molecular property prediction using pretrained-BERT and Bayesian active learning: a data-efficient approach to drug design

In drug discovery, prioritizing compounds for experimental testing is a critical task that can be optimized through active learning by strategically selecting informative molecules. Active learning typically trains models on labeled examples alone, while unlabeled data is only used for acquisition. This fully supervised approach neglects valuable information present in unlabeled molecular data, impairing both predictive performance and the molecule selection process. We address this limitation by integrating a transformer-based BERT model, pretrained on 1.26 million compounds, into the active learning pipeline. This effectively disentangles representation learning and uncertainty estimation, leading to more reliable molecule selection. Experiments on Tox21 and ClinTox datasets demonstrate that our approach achieves equivalent toxic compound identification with 50% fewer iterations compared to conventional active learning. Analysis reveals that pretrained BERT representations generate a structured embedding space enabling reliable uncertainty estimation despite limited labeled data, confirmed through Expected Calibration Error measurements. This work establishes that combining pretrained molecular representations with active learning significantly improves both model performance and acquisition efficiency in drug discovery, providing a scalable framework for compound prioritization. SCIENTIFIC CONTRIBUTION: We demonstrate that high-quality molecular representations fundamentally determine active learning success in drug discovery, outweighing acquisition strategy selection. We provide a framework that integrates pretrained transformer models with Bayesian active learning to separate representation learning from uncertainty estimation-a critical distinction in low-data scenarios. This approach establishes a foundation for more efficient screening workflows across diverse pharmaceutical applications.

GymBuddy and Elomia, AI-integrated applications, effects on the mental health of the students with psychological disorders

BACKGROUND

Digital mental health interventions, including AI-integrated applications, are increasingly utilized to support individuals with elevated symptoms of psychological distress. However, a gap exists in understanding their efficacy specifically for student populations.

OBJECTIVES

This study aimed to investigate the effects of GymBuddy, an AI-powered fitness and accountability app, and Elomia, an AI-based mental health chatbot, on the mental health of students at risk for psychological distress.

METHODOLOGY

A quasi-experimental study was conducted involving 65 participants who exhibited heightened psychological distress but did not have a formal diagnosis of a psychological disorder. Participants were randomly assigned to either the intervention group, which utilized GymBuddy and Elomia for structured mental health support, or the control group. Mental health outcomes such as anxiety, depression, and stress levels were assessed using standardized baseline, midpoint, and endpoint measures. Data were analyzed using Mixed ANOVA.

RESULTS

The mixed ANOVA analysis revealed significant improvements across all measured mental health outcomes, including somatic symptoms, anxiety and insomnia, social dysfunction, and severe depression. Significant main effects of time and group membership were observed for all variables, indicating overall symptom reduction and baseline differences between groups. Moreover, significant interaction effects for somatic symptoms (F(2, 70) = 59.96, p < 0.0001, η² = 0.63), anxiety and insomnia (F(2, 70) = 32.05, p < 0.0001, η² = 0.48), social dysfunction (F(2, 70) = 59.96, p < 0.0001, η² = 0.63), and severe depression (F(2, 70) = 32.05, p < 0.0001, η² = 0.48) indicated that participants in the intervention group experienced significantly greater reductions in psychological distress compared to the control group.

CONCLUSIONS

Our findings suggest that AI-integrated interventions like GymBuddy and Elomia may serve as effective tools for reducing psychological distress in student populations. Integrating AI technology into mental health interventions offers personalized support and guidance, addressing a crucial need in student populations. Further research is warranted to explore long-term outcomes and optimize the implementation of these interventions in educational settings.

MulAFNet: Integrating Multiple Molecular Representations for Enhanced Property Prediction

In computer-aided drug design, molecular representation plays a crucial role. Most existing multimodal approaches primarily perform simple concatenation of various feature representations, without adequately emphasizing effective integration among these features. To address this issue, this study proposes a network framework that integrates multimodal representations using a multihead attention flow (MulAFNet). MulAFNet utilizes SMILES string representation and two levels of molecular graph representations: atom-level and functional group-level graph structure. Pretraining tasks are established for each of these three representations, which are then fused in downstream tasks to predict molecular properties. The experiments were conducted on six classification data sets and three regression data sets, demonstrating that the use of multiple molecular representations as input has a significant impact on the results. In particular, the excellent performance of our fusion method in molecular property prediction outperforms other state-of-the-art methods, proving its superiority. Additionally, comparative experiments on fusion methods and ablation studies, further validate the effectiveness of MulAFNet. The results demonstrate that multiple molecular feature representations provide a more comprehensive molecular understanding, and appropriate pretraining tasks enhance molecular property prediction.

Artificial intelligence-based molecular property prediction of photosensitising effects of drugs

Drug-induced photosensitivity is a potential adverse event of many drugs and chemicals used across a wide range of specialties in clinical medicine. In the present study, we investigated the feasibility of predicting the photosensitising effects of drugs and chemical compounds via state-of-the-art artificial intelligence-based workflows. A dataset of 2200 drugs was used to train three distinct models (logistic regression, XGBoost and a deep learning model (Chemprop)) to predict photosensitising attributes. Labels were obtained from a list of previously published photosensitisers by string matching and manual validation. External evaluation of the different models was performed using the tox21 dataset. ROC-AUC ranged between 0.8939 (Chemprop) and 0.9525 (XGBoost) during training, while in the test partition it ranged between 0.7785 (Chemprop) and 0.7927 (XGBoost). Analysis of the top 200 compounds of each model resulted in 55 overlapping molecules in the external validation set. Prediction scores in fluoroquinolones within this subset corresponded well with culprit substructures such as fluorinated aryl halides suspected of mediating photosensitising effects. All three models appeared capable of predicting photosensitising effects of chemical compounds. However, compared to the simpler model, the complex models appeared to be more confident in their predictions as exhibited by their distribution of prediction scores.

Bridging imaging-based in vitro methods from biomedical research to regulatory toxicology

Imaging technologies are being increasingly used in biomedical research and experimental toxicology to gather morphological and functional information from cellular models. There is a concrete opportunity of incorporating imaging-based in vitro methods in international guidelines to respond to regulatory requirements with human relevant data. To translate these methods from R&D to international regulatory acceptance, the community needs to implement test methods under quality management systems, assess inter-laboratory transferability, and demonstrate data reliability and robustness. This article summarises current challenges associated with image acquisition, image analysis, including artificial intelligence, and data management of imaging-based methods, with examples from the developmental neurotoxicity in vitro battery and phenotypic profiling assays. The article includes considerations on specific needs and potential solutions to design and implement future validation and transferability studies.

Toxic Alerts of Endocrine Disruption Revealed by Explainable Artificial Intelligence

The local interpretable model-agnostic explanation method was used to unveil substructures (toxic alerts) that cause endocrine disruption in chemical compounds using machine learning models. The random forest classifier was applied to build explainable models with the TOX21 data sets after data curation. Using these models applied to the EDC and EDKB-FDA data sets, the substructures that cause endocrine disruption in chemical compounds were unveiled, providing stable, more specific, and consistent explanations, which are essential for trust and acceptance of the findings, mainly due to the difficulty of finding relevant experimental evidence for different receptors (androgen, estrogen, aryl hydrocarbon, aromatase, and peroxisome proliferator-activated receptors). This approach is significant because of its contribution to the interpretability of explainable machine learning algorithms, particularly in the context of unveiling substructures associated with endocrine disruption in five targets (androgen receptor, estrogen receptor, aryl hydrocarbon receptors, aromatase receptors, and peroxisome proliferator-activated receptors), thereby advancing the relevant field of environmental toxicology, where a careful evaluation of the potential risks of exposure to new compounds is needed. The specific substructures thiophosphate, sulfamate, anilide, carbamate, sulfamide, and thiocyanate are presented as toxic alerts that cause endocrine disruption to better understand their potential risks and adverse effects on human health and the environment.

Consensus Modeling for Predicting Chemical Binding to Transthyretin as the Winning Solution of the Tox24 Challenge

The utilization of predictive methodologies for the assessment of toxicological properties represents an alternative approach that facilitates the identification of safe compounds while concurrently reducing the financial costs associated with the process. The objective of the Tox24 Challenge was to assess the progress in computational methods for predicting the activity of chemical binding to transthyretin (TTR). In order to fulfill the requirements of this task, the data set, measured by the Environmental Protection Agency, consisted of 1512 chemical substances of diverse nature. This paper describes the model that won the Tox24 Challenge and the steps taken for its further improvement. The Transformer convolutional neural network (CNN) model achieved the best performance as a standalone solution. Meanwhile, a multitask model built on a graph CNN, trained using 11 additional acute systemic toxicity data sets with increased weighting on the TTR binding activity, showed comparable results on the blind test set. The winning solution was a consensus model consisting of two catBoost models with OEstate and Mold2 descriptor sets, as well as two transformer-based models. The improvement of this solution involved adding a fifth model based on multitask learning using the graph CNN method, which led to a reduction in RMSE on the blind test set to 20.3%. The winning model was developed using the OCHEM web platform and is available online at https://ochem.eu/article/162082.

From High Resolution Tandem Mass Spectrometry to Pollutant Toxicity AI-Based Prediction: A Case Study of 7 Endocrine Disruptors Endpoints

Based on high-resolution mass spectrometry (HRMS), nontarget analysis (NTA) can rapidly identify and characterize numerous hazardous substances in complex environmental samples. However, the intricate identification process often results in the underutilization of many mass spectrometry features. Even when chemical structures are identified, their toxicological effects and health outcomes may remain unknown. To address these challenges, this study introduces MSFragTox, a novel approach that leverages the rich fragmentation spectra inherent in high resolution tandem mass spectrometry (MS/MS) to directly predict toxicity. This method integrates MS/MS data with high-throughput screening (HTS) assays, focusing on seven endocrine disruption-related endpoints from Tox21, and uses MS-derived fingerprints: substructure fragmentation probability vectors to construct toxicity predictions using machine learning algorithms. The best model demonstrated robust performance with an average area under the receiver operating characteristic curve (AUROC) of 0.845 on the test set, outperforming models based on traditional molecular fingerprints and descriptors. Additionally, a web client (http://ms.envwind.site:8500) is provided for users to screen toxicity based on chemical MS/MS data. Furthermore, in-depth analyses of commonalities and differences in substructures reveal the mechanisms underlying across toxicity endpoints. Using MSFragTox, we validated the potential endocrine-disrupting effects of substances corresponding to MS/MS from real samples, highlighting the feasibility of directly studying toxicity through MS/MS and its potential applications in risk prediction and early warning for environmental samples.

Reference Library for Suspect Non-targeted Screening of Environmental Toxicants Using Ion Mobility Spectrometry-Mass Spectrometry

As our health is affected by the xenobiotic chemicals we are exposed to, it is important to rapidly assess these molecules both in the environment and our bodies. Targeted analytical methods coupling either gas or liquid chromatography with mass spectrometry (GC-MS or LC-MS) are commonly utilized in current exposure assessments. While these methods are accepted as the gold standard for exposure analyses, they often require multiple sample preparation steps and more than 30 minutes per sample. This throughput limitation is a critical gap for exposure assessments and has resulted in an evolving interest in using ion mobility spectrometry and MS (IMS-MS) for non-targeted studies. IMS-MS is a unique technique due to its rapid analytical capabilities (millisecond scanning) and detection of a wide range of chemicals based on unique collision cross section (CCS) and mass-to-charge (m/z) values. To increase the availability of IMS-MS information for exposure studies, here we utilized drift tube IMS-MS to evaluate 4,685 xenobiotic chemical standards from the Environmental Protection Agency Toxicity Forecaster (ToxCast) program including pesticides, industrial chemicals, pharmaceuticals, consumer products, and per- and polyfluoroalkyl substances (PFAS). In the analyses, 3,993 [M+H]+, [M+Na]+, [M-H]- and [M+]+ ion types were observed with high confidence and reproducibility (≤1% error intra-laboratory and ≤2% inter-laboratory) from 2,140 unique chemicals. These values were then assembled into an openly available multidimensional database and uploaded to PubChem to enable rapid IMS-MS suspect screening for a wide range of environmental contaminants, faster response time in environmental exposure assessments, and assessments of xenobiotic-disease connections.

SuperResNET - single-molecule network analysis detects changes to clathrin structure induced by small-molecule inhibitors

SuperResNET is a network analysis pipeline for the analysis of point cloud data generated by single-molecule localization microscopy (SMLM). Here, we applied SuperResNET network analysis of SMLM direct stochastic optical reconstruction microscopy (dSTORM) data to determine how the clathrin endocytosis inhibitors pitstop 2, dynasore and latrunculin A (LatA) alter the morphology of clathrin-coated pits. SuperResNET analysis of HeLa and Cos7 cells identified three classes of clathrin structures: small oligomers (class I), pits and vesicles (class II), and larger clusters corresponding to fused pits or clathrin plaques (class III). Pitstop 2 and dynasore treatment induced distinct homogeneous populations of class II structures in HeLa cells, suggesting that they arrest endocytosis at different stages. Inhibition of endocytosis was not via actin depolymerization, as the actin-depolymerizing agent LatA induced large, heterogeneous clathrin structures. Ternary analysis of SuperResNET shape features presented a distinct more planar profile for blobs from pitstop 2-treated cells, which aligned with clathrin pits identified with high-resolution minimal photon fluxes (MINFLUX) microscopy, whereas control structures resembled MINFLUX clathrin vesicles. SuperResNET analysis therefore showed that pitstop 2 arrests clathrin pit maturation at early stages of pit formation, representing an approach to detect the effect of small molecules on target structures in situ in the cell from SMLM datasets.

MLinvitroTox reloaded for high-throughput hazard-based prioritization of high-resolution mass spectrometry data

MLinvitroTox is an automated Python pipeline developed for high-throughput hazard-driven prioritization of toxicologically relevant signals detected in complex environmental samples through high-resolution tandem mass spectrometry (HRMS/MS). MLinvitroTox is a machine learning (ML) framework comprising 490 independent XGBoost classifiers trained on molecular fingerprints from chemical structures and target-specific endpoints from the ToxCast/Tox21 invitroDBv4.1 database. For each analyzed HRMS feature, MLinvitroTox generates a 490-bit bioactivity fingerprint used as a basis for prioritization, focusing the time-consuming molecular identification efforts on features most likely to cause adverse effects. The practical advantages of MLinvitroTox are demonstrated for groundwater HRMS data. Among the 874 features for which molecular fingerprints were derived from spectra, including 630 nontargets, 185 spectral matches, and 59 targets, around 4% of the feature/endpoint relationship pairs were predicted to be active. Cross-checking the predictions for targets and spectral matches with invitroDB data confirmed the bioactivity of 120 active and 6791 nonactive pairs while mislabeling 88 active and 56 non-active relationships. By filtering according to bioactivity probability, endpoint scores, and similarity to the training data, the number of potentially toxic features was reduced by at least one order of magnitude. This refinement makes the analytical confirmation of the toxicologically most relevant features feasible, offering significant benefits for cost-efficient chemical risk assessment.Scientific Contribution:In contrast to the classical ML-based approaches for toxicity prediction, MLinvitroTox predicts bioactivity for HRMS features (i.e., distinct m/z signals) based on MS2 fragmentation spectra rather than the chemical structures from the identified features. While the original proof of concept study was accompanied by the release of a MLinvitroTox v1 KNIME workflow, in this study, we release a Python MLinvitroTox v2 package, which, in addition to automation, expands functionality to include predicting toxicity from structures, cleaning up and generating chemical fingerprints, customizing models, and retraining on custom data. Furthermore, as a result of improvements in bioactivity data processing, realized in the concurrently released pytcpl Python package for the custom processing of invitroDBv4.1 input data used for training MLinvitroTox, the current release introduces enhancements in model accuracy, coverage of biological mechanistic targets, and overall interpretability.

Analytical Quality Evaluation of the Tox21 Compound Library

The analytical quality of compounds subjected to high-throughput screening (HTS) impacts accurate interpretation of assay results, with poor quality samples potentially leading to false negatives or positives. The Tox21 "10K" library consists of over 8900 unique compounds, spanning a diverse landscape of environmental and pharmaceutical chemicals, posing opportunities and challenges for analytical quality control (QC) determinations. Tox21 sample plates stored in DMSO at ambient conditions for 0 (T0) and/or 4 months (T4), totaling more than 13K unique sample identifiers (Tox21 IDs), were subjected to various analyses, including liquid and gas chromatography mass spectrometry (LC-MS, GC-MS) and nuclear magnetic resonance (NMR). Results for each sample at T0 or T4 underwent expert review and, where possible, a QC grade conveying purity, identity, and concentration was assigned. Herein, we relate details of the methods applied and report on the original (v0) Tox21 ID level results. Thirteen QC grades were condensed to 5 quality scores to aid global analysis, resulting in reinterpretation and improvement of >700 sample grades. Of the 92% T0 samples successfully graded, 76% exceeded 90% purity. For 76% of samples that were also tested at T4, 89% showed no evidence of sample loss or degradation. Prioritized quality bins were used to summarize thousands of replicate sample-level QC results to a compound-level QC score to support structure-based analyses. ToxPrint chemotype analysis identified structural features enriched in unstable compounds, as well as in high and low quality T0 subsets. Predicted vapor pressure was weakly correlated with low-concentration QC indicators, reflecting likely entanglement with method amenability and quality issues. Finally, an ongoing EPA effort to re-evaluate the original QC spectra is generating insights that will further modify QC grades. Tox21 QC spectra and results will be made available in a new public QC browser, facilitating further evaluation to support HTS interpretation and modeling applications.

Unique and Common Agonists Activate the Insect Juvenile Hormone Receptor and the Human AHR

Transcription factors of the bHLH-PAS family play vital roles in animal development, physiology, and disease. Two members of the family require binding of low-molecular weight ligands for their activity: the vertebrate aryl hydrocarbon receptor (AHR) and the insect juvenile hormone receptor (JHR). In the fly Drosophila melanogaster, the paralogous proteins GCE and MET constitute the ligand-binding component of JHR complexes. Whilst GCE/MET and AHR are phylogenetically heterologous, their mode of action is similar. JHR is targeted by several synthetic agonists that serve as insecticides disrupting the insect endocrine system. AHR is an important regulator of human endocrine homeostasis, and it responds to environmental pollutants and endocrine disruptors. Whether AHR signaling is affected by compounds that can activate JHR has not been reported. To address this question, we screened a chemical library of 50,000 compounds to identify 93 novel JHR agonists in a reporter system based on Drosophila cells. Of these compounds, 26% modulated AHR signaling in an analogous reporter assay in a human cell line, indicating a significant overlap in the agonist repertoires of the two receptors. To explore the structural features of agonist-dependent activation of JHR and AHR, we compared the ligand-binding cavities and their interactions with selective and common ligands of AHR and GCE. Molecular dynamics modeling revealed ligand-specific as well as conserved side chains within the respective cavities. Significance of predicted interactions was supported through site-directed mutagenesis. The results have indicated that synthetic insect juvenile hormone agonists might interfere with AHR signaling in human cells.

Coverage bias in small molecule machine learning

Small molecule machine learning aims to predict chemical, biochemical, or biological properties from molecular structures, with applications such as toxicity prediction, ligand binding, and pharmacokinetics. A recent trend is developing end-to-end models that avoid explicit domain knowledge. These models assume no coverage bias in training and evaluation data, meaning the data are representative of the true distribution. However, the domain of applicability is rarely considered in such models. Here, we investigate how well large-scale datasets cover the space of known biomolecular structures. For doing so, we propose a distance measure based on solving the Maximum Common Edge Subgraph (MCES) problem, which aligns well with chemical similarity. Although this method is computationally hard, we introduce an efficient approach combining Integer Linear Programming and heuristic bounds. Our findings reveal that many widely-used datasets lack uniform coverage of biomolecular structures, limiting the predictive power of models trained on them. We propose two additional methods to assess whether training datasets diverge from known molecular distributions, potentially guiding future dataset creation to improve model performance.

Attention-based deep learning for accurate cell image analysis

High-content analysis (HCA) holds enormous potential for drug discovery and research, but widely used methods can be cumbersome and yield inaccurate results. Noisy and redundant signals in cell images impede accurate deep learning-based image analysis. To address these issues, we introduce X-Profiler, a novel HCA method that combines cellular experiments, image processing, and deep learning modeling. X-Profiler combines the convolutional neural network and Transformer to encode high-content images, effectively filtering out noisy signals and precisely characterizing cell phenotypes. In comparative tests on drug-induced cardiotoxicity, mitochondrial toxicity classification, and compound classification, X-Profiler outperformed both DeepProfiler and CellProfiler, as two highly recognized and representative methods in this field. Our results demonstrate the utility and versatility of X-Profiler, and we anticipate its wide application in HCA for advancing drug development and disease research.

QSAR: Using the Past to Study the Present

Quantitative structure-activity relationships (QSAR) is a method for predicting the physical and biological properties of small molecules; it is in use in industry and public services. However, as any scientific method, it is challenged by more and more requests, especially considering its possible role in assessing the safety of new chemicals. To answer the question whether QSAR, by exploiting available knowledge, can build new knowledge, the chapter reviews QSAR methods in search of a QSAR epistemology. QSAR stands on tree pillars, i.e., biological data, chemical knowledge, and modeling algorithms. Usually the biological data, resulting from good experimental practice, are taken as a true picture of the world; chemical knowledge has scientific bases; so if a QSAR model is not working, blame modeling. The role of modeling in developing scientific theories, and in producing knowledge, is so analyzed. QSAR is a mature technology and is part of a large body of in silico methods and other computational methods. The active debate about the acceptability of the QSAR models, about the way to communicate them, and the explanation to provide accompanies the development of today QSAR models. An example about predicting possible endocrine-disrupting chemicals (EDC) shows the many faces of modern QSAR methods.

A weight of evidence review on the mode of action, adversity, and the human relevance of xylene's observed thyroid effects in rats

Xylene substances have wide industrial and consumer uses and are currently undergoing dossier and substance evaluation under Registration, Evaluation, Authorization and Restriction of Chemicals (REACH) for further toxicological testing including consideration of an additional neurotoxicological testing cohort to an extended one-generation reproduction toxicity (EOGRT) study. New repeated dose study data on xylenes identify the thyroid as a potential target tissue, and therefore a weight of evidence review is provided to investigate whether or not xylene-mediated changes on the hypothalamus-pituitary-thyroid (HPT) axis are secondary to liver enzymatic induction and are of a magnitude that is relevant for neurological human health concerns. Multiple published studies confirm xylene-mediated increases in liver weight, hepatocellular hypertrophy, and liver enzymatic induction via the oral or inhalation routes, including an increase in uridine 5'-diphospho-glucuronosyltransferase (UDP-GT) activity, the key step in thyroid hormone metabolism in rodents. Only minimal to slight increases in thyroid follicular cell hypertrophy have been observed in some xylene repeated dose studies, with no associated robust or consistent perturbance of thyroid hormone changes across the studies or carried through to offspring indicating adaptive homeostatic maintenance of the HPT axis. Also importantly, in vitro human cell line data from the United States Environmental Protection Agency (US EPA) Toxicity Forecasting (ToxCast) provides supporting evidence of xylene's inability to directly perturb thyroidal functionality. A further supplemental in-depth metabolomics analysis (MetaMap®Tox) of xylene showed a tentative match to compounds that also demonstrate extra-thyroidal effects on the HPT axis as a consequence of liver enzyme induction. Lastly, the slight HPT axis changes mediated by xylene were well-below the published literature thresholds for developmental neurotoxicological outcomes established for thyroidal changes in animals and humans. In summary, the data and various lines of scientific evidence presented herein individually and collectively demonstrate that xylene's mediated changes in the HPT axis, via a secondary extra-thyroidal MOA (i.e. liver enzyme induction), do not raise a human health concern with regards to developmental neurotoxicity. As such, the available toxicological data do not support the classification of xylene as a known or suspected endocrine disruptor, specifically through the thyroid modality, per Regulations Commission Delegated Regulation (EU) 2023/707 of 19 December 2022 amending Regulation (EC) No 1272/2008 and do not support the need for a neurotoxicological cohort evaluation in any subsequent EOGRTS.

In silico models for the screening of human transthyretin disruptors

The use of New Approach Methodologies (NAMs), such as Quantitative Structure-Activity Relationship (QSAR) models, is highly recommended by international regulations to speed up hazard and risk assessment of Endocrine Disruptors, which are known to be linked to a wide spectrum of severe diseases on humans and wildlife. A very sensitive target for these chemicals is the thyroid hormone system, which plays a key role in regulating metabolic and cognitive functions. Several chemicals have been demonstrated to compete with the thyroid hormone thyroxine (T4) for binding to human thyroid hormone distributor protein transthyretin (hTTR). In this work, we generated three new datasets composed by T4-hTTR competing potencies of more than 200 heterogeneous chemicals measured by three different in vitro assays. These datasets were used for the development of new regression QSAR models. The best models were thoroughly validated by internal and external validation procedures. The mechanistic interpretation of the selected molecular descriptors provided information on structural features which are relevant to characterise hTTR binders, such as the presence of hydroxylated and halogenated aromatic rings. PCA analysis was used to rank the studied chemicals according to their increasing T4-hTTR competing potency. Hydroxylated and halogenated bicyclic aromatic compounds are ranked as the strongest hTTR binders. The new QSARs are useful to screen potential Thyroid Hormone System-Disrupting Chemicals (THSDCs), and to support the identification of sustainable alternatives to hazardous chemicals.

Comprehensive Characterization of Oxidative Stress-Modulating Chemicals Using GPT-Based Text Mining

The screening of hazardous environmental pollutants is hindered by the limited availability of toxicological databases. Large language model (LLM)-based text mining holds the potential to automatically extract complex toxicological information from the literature. Due to its relevance to diseases and the challenge of comprehensive characterization, oxidative stress serves as a suitable case for research by texting mining. In this study, a robust workflow utilizing a LLM (i.e., GPT-4) was developed to extract information on oxidative stress tests, including data collection, text preprocessing, prompt engineering, and performance evaluation procedures. A total of 17,780 relevant records were extracted from 7166 articles, covering 2558 unique compounds. A rising interest in oxidative stress was observed over the past two decades. A list of known prooxidants (n = 1416) and antioxidants (n = 1102) was established, with the leading chemical categories being pharmaceuticals, pesticides, and metals for prooxidants and pharmaceuticals and flavonoids for antioxidants. Structural alert analysis identified potential prooxidant (e.g., chlorobenzene, nitrobenzene, and tertiary amines) and antioxidant (e.g., flavonoid and thiol) substructures. These findings illustrate the feasibility of building toxicological databases through LLM-based text mining in a cost-efficient manner, and the information obtained from the technique holds significant promise for future applications in environmental and health research.

Mechanism-based toxicity screening of organophosphate flame retardants using Tox21 assays and molecular docking analysis

As brominated flame retardants are phased out and regulations on their use become stricter, concerns over organophosphate flame retardants (OPFRs) have increased due to their high production. In response, this study aimed to screen the potential toxicity of emerging OPFRs using in vitro Tox21 assays and in silico molecular docking analysis. For 48 OPFRs collected from the literature, we investigated their bioactivity with human nuclear receptors using Tox21 data, focusing on pathways related to endocrine disruption (ERs, AR), stress response (GR), energy homeostasis (PPARs, FXR), and detoxification (PXR, CAR). For OPFRs not tested in Tox21 assays, molecular docking simulations were performed to predict binding potential. Results showed that CAR/PXR and FXR had relatively high reactivity with diverse OPFRs, indicating potential molecular initiating events (MIEs). Among the 48 OPFRs, 28 interacted with one or more receptors, suggesting they may act as potential stressors of adverse outcome pathways (AOPs) leading to various human diseases. Aryl- and halogenated-OPFRs displayed higher bioactivity compared to alkyl-OPFRs. Additionally, as the logKow value and carbon number of OPFRs increased, their interaction with nuclear receptors also increased. These structure- and physicochemistry-dependent bioactivities provide insights for designing safer OPFRs to avoid regrettable substitutions. Of these prioritized OPFRs, 13 showed low oral points-of-departure (POD) values under 100 mg/kg/day. In contrast, the other 15 OPFRs lacked sufficient data or exhibited less severe toxicity, despite being predicted to be of high concern in our analysis. Since several OPFRs are commonly used in consumer products that can lead to daily human exposure, we suggest that these OPFRs have the potential to reveal undisclosed effects and should therefore undergo further assessment.

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 10K 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 10K 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.

Evaluating the Utility of the MSTI Assay in Predicting Compound Promiscuity and Cytotoxicity

Nonspecific reactive chemicals often interfere with the interpretation of high-throughput assay results because of their promiscuity and/or cytotoxicity. Using a high-throughput assay to identify such compounds is necessary to efficiently rule out potential assay artifacts. The MSTI, (E)-2-(4-mercaptostyryl)-1,3,3-trimethyl-3H-indol-1-ium, assay uses a thiol-containing fluorescent probe to screen for electrophile reactivity and could potentially be used to determine nonspecific reactive compounds. The Tox21 10K compound library was previously screened against a panel of ∼80 cell-based and biochemical assays, including the biochemical MSTI assay. In this study, we compared the MSTI assay activity of the Tox21 10K compounds with their promiscuity and cytotoxicity as reflected by their activities across the Tox21 assay panel to determine: (1) if this assay is predictive of a compound's promiscuity and cytotoxicity and (2) what chemical features create inconsistent results between the MSTI assay activity and promiscuity/cytotoxicity (false negatives and false positives). We found that the MSTI assay can predict a chemical's promiscuity/cytotoxicity with a 0.55 sensitivity and 0.97 specificity. Out of 3,407 unique compounds evaluated, we identified 92 false positive and 227 false negative results. Several structural features such as carboxamides and alkyl halides were found to be apparent in 53% (p = 2.4 × 10-07) and 19% (p = 4.3 × 10-06) of the false positives and negatives, respectively. The results of this analysis will help identify the potential challenges of this high-throughput assay and allow researchers to identify if a compound will be cytotoxic or promiscuous in an efficient manner.

In Silico Prediction of ERRα Agonists Based on Combined Features and Stacking Ensemble Method

Estrogen-related receptor α (ERRα) is considered a very promising target for treating metabolic diseases such as type 2 diabetes. Development of a prediction model to quickly identify potential ERRα agonists can significantly reduce the time spent on virtual screening. In this study, 298 ERRα agonists and numerous nonagonists were collected from various sources to build a new dataset of ERRα agonists. Then a total of 90 models were built using a combination of different algorithms, molecular characterization methods, and data sampling techniques. The consensus model with optimal performance was also validated on the test set (AUC=0.876, BA=0.816) and external validation set (AUC=0.867, BA=0.777) based on five selected baseline models. Furthermore, the model's applicability domain and privileged substructures were examined, and the feature importance was analyzed using the SHAP method to help interpret the model. Based on the above, it's hoped that our publicly accessible data, models, codes, and analytical techniques will prove valuable in quick screening and rational designing more novel and potent ERRα agonists.

Identifying long-term health risks associated with environmental chemical incidents

In recent years, there has been a notable surge in environmental incidents, including wildfires and chemical releases. Responses to such events have primarily focused on addressing acute and immediate impacts. However, potential long-term health risks have been overlooked. Our proposed framework first advocates for the holistic identification of contaminants, prioritizing persistent organic contaminants determined through both knowledge-based and non-targeted and targeted analysis. We suggest integrating environmental monitoring and modeling approaches to assess the extent and composition of contamination caused by these chemicals. To facilitate swift assessments, we advocate the development of streamlined chemical analysis techniques and dedicated technologies for in situ monitoring of persistent organic chemicals. In addition, we provide an overview of both traditional and state-of-the-art approaches to risk assessment and introduce a three-tier risk assessment framework for evaluating the long-term health risks associated with environmental incidents. We emphasize the importance of in situ soil remediation and coordinated recovery efforts, including effective communication, evacuation, and cleaning plans for affected spaces, which are pivotal for facilitating recovery from environmental incidents. This comprehensive approach fortifies preparedness and recovery strategies, providing a robust framework for managing future environmental crises.

Emerging investigator series: nontargeted screening of aryl hydrocarbon receptor agonists in endangered beluga whales from the St. Lawrence Estuary: beyond legacy contaminants

The elevated concentrations of organohalogen contaminants in the endangered St. Lawrence Estuary (SLE) belugas have prompted the hypothesis that aryl hydrocarbon receptor (AhR) activity may be a contributor towards their potential adverse effects. While indirect associations between AhR and contaminant levels have been reported in SLE beluga tissues, AhR activity was never directly measured. Using bioassays and nontargeted analysis, this study contrasted AhR activity and agonist profiles between pooled tissue extracts of endangered SLE and non-threatened Arctic belugas. Tissue extracts of SLE belugas exhibited significantly higher overall AhR activity than that of Arctic belugas, with a 2000s SLE beluga liver extract exerting significantly higher activity than blubber extracts of SLE and Arctic belugas from the same time period. Contrary to our expectations, well-known AhR agonists detected by nontargeted analysis, including polychlorinated biphenyls (PCBs), were only minor contributors to the observed AhR activity. Instead, Tox21 suspect screening identified more polar chemicals, such as dyes and natural indoles, as potential contributors. Notably, the natural product bromoindole was selectively detected in SLE beluga liver at high abundance and was further confirmed as an AhR agonist. These findings highlighted the significance of the AhR-mediated toxicity pathway in belugas and underscored the importance of novel AhR agonists, particularly polar compounds, in its induction.

Knowledge-based machine learning for predicting and understanding the androgen receptor (AR)-mediated reproductive toxicity in zebrafish

Traditional methods for identifying endocrine-disrupting chemicals (EDCs) that activate androgen receptors (AR) are costly, time-consuming, and low-throughput. This study developed a knowledge-based deep neural network model (AR-DNN) to predict AR-mediated adverse outcomes on female zebrafish fertility. This model started with chemical fingerprints as the input layer and was implemented through a five-layer virtual AR-induced adverse outcome pathway (AOP). Results indicated that the AR-DNN effectively and accurately screens new reproductive toxicants (AUC = 0.94, accuracy = 0.85), providing potential toxicity pathways. Furthermore, 1477 and 2448 chemicals that could lead to infertility were identified in the plastic additives list (PLASTICMAP, n = 7112) and the Inventory of Existing Chemical Substances in China (IECSC, n = 17741), respectively. Colourants containing steroid-like structures are the major active plastic additives that might lower female zebrafish fertility through AR binding, DNA binding, and transcriptional activation. While active IECSC chemicals primarily have the same fragments, such as benzonitrile, nitrobenzene, and quinolone. The predicted toxicity pathways were consistent with existing fish evidence, demonstrating the model's applicability. This knowledge-based approach offers a promising computational toxicology strategy for predicting and characterising the endocrine-disrupting effects and toxic mechanisms of organic chemicals, potentially leading to more efficient and cost-effective screening of EDCs.

Identification of Environmental Compounds That May Trigger Early Female Puberty by Activating Human GnRHR and KISS1R

There has been an alarming trend toward earlier puberty in girls, suggesting the influence of an environmental factor(s). As the reactivation of the reproductive axis during puberty is thought to be mediated by the hypothalamic neuropeptides kisspeptin and gonadotropin-releasing hormone (GnRH), we asked whether an environmental compound might activate the kisspeptin (KISS1R) or GnRH receptor (GnRHR). We used GnRHR or KISS1R-expressing HEK293 cells to screen the Tox21 10K compound library, a compendium of pharmaceuticals and environmental compounds, for GnRHR and KISS1R activation. Agonists were identified using Ca2+ flux and phosphorylated extracellularly regulated kinase (p-ERK) detection assays. Follow-up studies included measurement of genes known to be upregulated upon receptor activation using relevant murine or human cell lines and molecular docking simulation. Musk ambrette was identified as a KISS1R agonist, and treatment with musk ambrette led to increased expression of Gnrh1 in murine and human hypothalamic cells and expansion of GnRH neuronal area in developing zebrafish larvae. Molecular docking demonstrated that musk ambrette interacts with the His309, Gln122, and Gln123 residues of the KISS1R. A group of cholinergic agonists with structures similar to methacholine was identified as GnRHR agonists. When applied to murine gonadotrope cells, these agonists upregulated Fos, Jun, and/or Egr1. Molecular docking revealed a potential interaction between GnRHR and 5 agonists, with Asn305 constituting the most conservative GnRHR binding site. In summary, using a Tox21 10K compound library screen combined with cellular, molecular, and structural biology techniques, we have identified novel environmental agents that may activate the human KISS1R or GnRHR.

Cheminformatic Read-Across Approach Revealed Ultraviolet Filter Cinoxate as an Obesogenic Peroxisome Proliferator-Activated Receptor γ Agonist

This study introduces a novel cheminformatic read-across approach designed to identify potential environmental obesogens, substances capable of disrupting metabolism and inducing obesity by mainly influencing nuclear hormone receptors (NRs). Leveraging real-valued two-dimensional features derived from chemical fingerprints of 8435 Tox21 compounds, cluster analysis and subsequent statistical testing revealed 385 clusters enriched with compounds associated with specific NR targets. Notably, one cluster exhibited selective enrichment in peroxisome proliferator-activated receptor γ (PPARγ) agonist activity, prominently featuring methoxy cinnamate ultraviolet (UV) filters and obesogen-related compounds. Experimental validation confirmed that 2-ethoxyethyl 4-methoxycinnamate, an organic UV filter cinoxate, could selectively bind to PPARγ (Ki = 18.0 μM), eliciting an obesogenic phenotype in human bone marrow-derived mesenchymal stem cells during adipogenic differentiation. Molecular docking and further experiments identified cinoxate as a potent PPARγ full agonist, demonstrating a preference for coactivator SRC3 recruitment. Moreover, cinoxate upregulated transcription levels of genes encoding lipid metabolic enzymes in normal human epidermal keratinocytes as primary cells exposed during clinical usage. This study provides compelling evidence for the efficacy of cheminformatic read-across analysis in prioritizing potential obesogens, showcasing its utility in unveiling cinoxate as an obesogenic PPARγ agonist.

Moldable Plastics (Polycaprolactone) can be Acutely Toxic to Developing Zebrafish and Activate Nuclear Receptors in Mammalian Cells

Popularized on social media, hand-moldable plastics are formed by consumers into tools, trinkets, and dental prosthetics. Despite the anticipated dermal and oral contact, manufacturers share little information with consumers about these materials, which are typically sold as microplastic-sized resin pellets. Inherent to their function, moldable plastics pose a risk of dermal and oral exposure to unknown leachable substances. We analyzed 12 moldable plastics advertised for modeling and dental applications and determined them to be polycaprolactone (PCL) or thermoplastic polyurethane (TPU). The bioactivities of the most popular brands advertised for modeling applications of each type of polymer were evaluated using a zebrafish embryo bioassay. While water-borne exposure to the TPU pellets did not affect the targeted developmental end points at any concentration tested, the PCL pellets were acutely toxic above 1 pellet/mL. The aqueous leachates of the PCL pellets demonstrated similar toxicity. Methanolic extracts from the PCL pellets were assayed for their bioactivity using the Attagene FACTORIAL platform. Of the 69 measured end points, the extracts activated nuclear receptors and transcription factors for xenobiotic metabolism (pregnane X receptor, PXR), lipid metabolism (peroxisome proliferator-activated receptor γ, PPARγ), and oxidative stress (nuclear factor erythroid 2-related factor 2, NRF2). By nontargeted high-resolution comprehensive two-dimensional gas chromatography (GC × GC-HRT), we tentatively identified several compounds in the methanolic extracts, including PCL oligomers, a phenolic antioxidant, and residues of suspected antihydrolysis and cross-linking additives. In a follow-up zebrafish embryo bioassay, because of its stated high purity, biomedical grade PCL was tested to mitigate any confounding effects due to chemical additives in the PCL pellets; it elicited comparable acute toxicity. From these orthogonal and complementary experiments, we suggest that the toxicity was due to oligomers and nanoplastics released from the PCL rather than chemical additives. These results challenge the perceived and assumed inertness of plastics and highlight their multiple sources of toxicity.

Artificial intelligence and cheminformatics tools: a contribution to the drug development and chemical science

In the ever-evolving field of drug discovery, the integration of Artificial Intelligence (AI) and Machine Learning (ML) with cheminformatics has proven to be a powerful combination. Cheminformatics, which combines the principles of computer science and chemistry, is used to extract chemical information and search compound databases, while the application of AI and ML allows for the identification of potential hit compounds, optimization of synthesis routes, and prediction of drug efficacy and toxicity. This collaborative approach has led to the discovery, preclinical evaluations and approval of over 70 drugs in recent years. To aid researchers in the pursuit of new drugs, this article presents a comprehensive list of databases, datasets, predictive and generative models, scoring functions and web platforms that have been launched between 2021 and 2022. These resources provide a wealth of information and tools for computer-assisted drug development, and are a valuable asset for those working in the field of cheminformatics. Overall, the integration of AI, ML and cheminformatics has greatly advanced the drug discovery process and continues to hold great potential for the future. As new resources and technologies become available, we can expect to see even more groundbreaking discoveries and advancements in these fields.Communicated by Ramaswamy H. Sarma.

Deep learning algorithms applied to computational chemistry

Recently, there has been a significant increase in the use of deep learning techniques in the molecular sciences, which have shown high performance on datasets and the ability to generalize across data. However, no model has achieved perfect performance in solving all problems, and the pros and cons of each approach remain unclear to those new to the field. Therefore, this paper aims to review deep learning algorithms that have been applied to solve molecular challenges in computational chemistry. We proposed a comprehensive categorization that encompasses two primary approaches; conventional deep learning and geometric deep learning models. This classification takes into account the distinct techniques employed by the algorithms within each approach. We present an up-to-date analysis of these algorithms, emphasizing their key features and open issues. This includes details of input descriptors, datasets used, open-source code availability, task solutions, and actual research applications, focusing on general applications rather than specific ones such as drug discovery. Furthermore, our report discusses trends and future directions in molecular algorithm design, including the input descriptors used for each deep learning model, GPU usage, training and forward processing time, model parameters, the most commonly used datasets, libraries, and optimization schemes. This information aids in identifying the most suitable algorithms for a given task. It also serves as a reference for the datasets and input data frequently used for each algorithm technique. In addition, it provides insights into the benefits and open issues of each technique, and supports the development of novel computational chemistry systems.

High-Resolution Mass Spectrometry for Human Exposomics: Expanding Chemical Space Coverage

In the modern "omics" era, measurement of the human exposome is a critical missing link between genetic drivers and disease outcomes. High-resolution mass spectrometry (HRMS), routinely used in proteomics and metabolomics, has emerged as a leading technology to broadly profile chemical exposure agents and related biomolecules for accurate mass measurement, high sensitivity, rapid data acquisition, and increased resolution of chemical space. Non-targeted approaches are increasingly accessible, supporting a shift from conventional hypothesis-driven, quantitation-centric targeted analyses toward data-driven, hypothesis-generating chemical exposome-wide profiling. However, HRMS-based exposomics encounters unique challenges. New analytical and computational infrastructures are needed to expand the analysis coverage through streamlined, scalable, and harmonized workflows and data pipelines that permit longitudinal chemical exposome tracking, retrospective validation, and multi-omics integration for meaningful health-oriented inferences. In this article, we survey the literature on state-of-the-art HRMS-based technologies, review current analytical workflows and informatic pipelines, and provide an up-to-date reference on exposomic approaches for chemists, toxicologists, epidemiologists, care providers, and stakeholders in health sciences and medicine. We propose efforts to benchmark fit-for-purpose platforms for expanding coverage of chemical space, including gas/liquid chromatography-HRMS (GC-HRMS and LC-HRMS), and discuss opportunities, challenges, and strategies to advance the burgeoning field of the exposome.

A Chemical Structure and Machine Learning Approach to Assess the Potential Bioactivity of Endogenous Metabolites and Their Association with Early Childhood Systemic Inflammation

Metabolomics has gained much attention due to its potential to reveal molecular disease mechanisms and present viable biomarkers. This work uses a panel of untargeted serum metabolomes from 602 children from the COPSAC2010 mother-child cohort. The annotated part of the metabolome consists of 517 chemical compounds curated using automated procedures. We created a filtering method for the quantified metabolites using predicted quantitative structure-bioactivity relationships for the Tox21 database on nuclear receptors and stress response in cell lines. The metabolites measured in the children's serums are predicted to affect specific targeted models, known for their significance in inflammation, immune function, and health outcomes. The targets from Tox21 have been used as targets with quantitative structure-activity relationships (QSARs). They were trained for ~7000 structures, saved as models, and then applied to the annotated metabolites to predict their potential bioactivities. The models were selected based on strict accuracy criteria surpassing random effects. After application, 52 metabolites showed potential bioactivity based on structural similarity with known active compounds from the Tox21 set. The filtered compounds were subsequently used and weighted by their bioactive potential to show an association with early childhood hs-CRP levels at six months in a linear model supporting a physiological adverse effect on systemic low-grade inflammation.

Recent advances and current challenges of new approach methodologies in developmental and adult neurotoxicity testing

Adult neurotoxicity (ANT) and developmental neurotoxicity (DNT) assessments aim to understand the adverse effects and underlying mechanisms of toxicants on the human nervous system. In recent years, there has been an increasing focus on the so-called new approach methodologies (NAMs). The Organization for Economic Co-operation and Development (OECD), together with European and American regulatory agencies, promote the use of validated alternative test systems, but to date, guidelines for regulatory DNT and ANT assessment rely primarily on classical animal testing. Alternative methods include both non-animal approaches and test systems on non-vertebrates (e.g., nematodes) or non-mammals (e.g., fish). Therefore, this review summarizes the recent advances of NAMs focusing on ANT and DNT and highlights the potential and current critical issues for the full implementation of these methods in the future. The status of the DNT in vitro battery (DNT IVB) is also reviewed as a first step of NAMs for the assessment of neurotoxicity in the regulatory context. Critical issues such as (i) the need for test batteries and method integration (from in silico and in vitro to in vivo alternatives, e.g., zebrafish, C. elegans) requiring interdisciplinarity to manage complexity, (ii) interlaboratory transferability, and (iii) the urgent need for method validation are discussed.

Predicting the Activity of Unidentified Chemicals in Complementary Bioassays from the HRMS Data to Pinpoint Potential Endocrine Disruptors

The majority of chemicals detected via nontarget liquid chromatography high-resolution mass spectrometry (HRMS) in environmental samples remain unidentified, challenging the capability of existing machine learning models to pinpoint potential endocrine disruptors (EDs). Here, we predict the activity of unidentified chemicals across 12 bioassays related to EDs within the Tox21 10K dataset. Single- and multi-output models, utilizing various machine learning algorithms and molecular fingerprint features as an input, were trained for this purpose. To evaluate the models under near real-world conditions, Monte Carlo sampling was implemented for the first time. This technique enables the use of probabilistic fingerprint features derived from the experimental HRMS data with SIRIUS+CSI:FingerID as an input for models trained on true binary fingerprint features. Depending on the bioassay, the lowest false-positive rate at 90% recall ranged from 0.251 (sr.mmp, mitochondrial membrane potential) to 0.824 (nr.ar, androgen receptor), which is consistent with the trends observed in the models' performances submitted for the Tox21 Data Challenge. These findings underscore the informativeness of fingerprint features that can be compiled from HRMS in predicting the endocrine-disrupting activity. Moreover, an in-depth SHapley Additive exPlanations analysis unveiled the models' ability to pinpoint structural patterns linked to the modes of action of active chemicals. Despite the superior performance of the single-output models compared to that of the multi-output models, the latter's potential cannot be disregarded for similar tasks in the field of in silico toxicology. This study presents a significant advancement in identifying potentially toxic chemicals within complex mixtures without unambiguous identification and effectively reducing the workload for postprocessing by up to 75% in nontarget HRMS.

Chemistry domain of applicability evaluation against existing estrogen receptor high-throughput assay-based activity models

Introduction

The U. S. Environmental Protection Agency's Endocrine Disruptor Screening Program (EDSP) Tier 1 assays are used to screen for potential endocrine system-disrupting chemicals. A model integrating data from 16 high-throughput screening assays to predict estrogen receptor (ER) agonism has been proposed as an alternative to some low-throughput Tier 1 assays. Later work demonstrated that as few as four assays could replicate the ER agonism predictions from the full model with 98% sensitivity and 92% specificity. The current study utilized chemical clustering to illustrate the coverage of the EDSP Universe of Chemicals (UoC) tested in the existing ER pathway models and to investigate the utility of chemical clustering to evaluate the screening approach using an existing 4-assay model as a test case. Although the full original assay battery is no longer available, the demonstrated contribution of chemical clustering is broadly applicable to assay sets, chemical inventories, and models, and the data analysis used can also be applied to future evaluation of minimal assay models for consideration in screening.

Methods

Chemical structures were collected for 6,947 substances via the CompTox Chemicals Dashboard from the over 10,000 UoC and grouped based on structural similarity, generating 826 chemical clusters. Of the 1,812 substances run in the original ER model, 1,730 substances had a single, clearly defined structure. The ER model chemicals with a clearly defined structure that were not present in the EDSP UoC were assigned to chemical clusters using a k-nearest neighbors approach, resulting in 557 EDSP UoC clusters containing at least one ER model chemical.

Results and Discussion

Performance of an existing 4-assay model in comparison with the existing full ER agonist model was analyzed as related to chemical clustering. This was a case study, and a similar analysis can be performed with any subset model in which the same chemicals (or subset of chemicals) are screened. Of the 365 clusters containing >1 ER model chemical, 321 did not have any chemicals predicted to be agonists by the full ER agonist model. The best 4-assay subset ER agonist model disagreed with the full ER agonist model by predicting agonist activity for 122 chemicals from 91 of the 321 clusters. There were 44 clusters with at least two chemicals and at least one agonist based upon the full ER agonist model, which allowed accuracy predictions on a per-cluster basis. The accuracy of the best 4-assay subset ER agonist model ranged from 50% to 100% across these 44 clusters, with 32 clusters having accuracy ≥90%. Overall, the best 4-assay subset ER agonist model resulted in 122 false-positive and only 2 false-negative predictions compared with the full ER agonist model. Most false positives (89) were active in only two of the four assays, whereas all but 11 true positive chemicals were active in at least three assays. False positive chemicals also tended to have lower area under the curve (AUC) values, with 110 out of 122 false positives having an AUC value below 0.214, which is lower than 75% of the positives as predicted by the full ER agonist model. Many false positives demonstrated borderline activity. The median AUC value for the 122 false positives from the best 4-assay subset ER agonist model was 0.138, whereas the threshold for an active prediction is 0.1.

Conclusion

Our results show that the existing 4-assay model performs well across a range of structurally diverse chemicals. Although this is a descriptive analysis of previous results, several concepts can be applied to any screening model used in the future. First, the clustering of the chemicals provides a means of ensuring that future screening evaluations consider the broad chemical space represented by the EDSP UoC. The clusters can also assist in prioritizing future chemicals for screening in specific clusters based on the activity of known chemicals in those clusters. The clustering approach can be useful in providing a framework to evaluate which portions of the EDSP UoC chemical space are reliably covered by in silico and in vitro approaches and where predictions from either method alone or both methods combined are most reliable. The lessons learned from this case study can be easily applied to future evaluations of model applicability and screening to evaluate future datasets.

Expanding Predictive Capacities in Toxicology: Insights from Hackathon-Enhanced Data and Model Aggregation

In the realm of predictive toxicology for small molecules, the applicability domain of QSAR models is often limited by the coverage of the chemical space in the training set. Consequently, classical models fail to provide reliable predictions for wide classes of molecules. However, the emergence of innovative data collection methods such as intensive hackathons have promise to quickly expand the available chemical space for model construction. Combined with algorithmic refinement methods, these tools can address the challenges of toxicity prediction, enhancing both the robustness and applicability of the corresponding models. This study aimed to investigate the roles of gradient boosting and strategic data aggregation in enhancing the predictivity ability of models for the toxicity of small organic molecules. We focused on evaluating the impact of incorporating fragment features and expanding the chemical space, facilitated by a comprehensive dataset procured in an open hackathon. We used gradient boosting techniques, accounting for critical features such as the structural fragments or functional groups often associated with manifestations of toxicity.

In Silico and In Vitro multiple analysis approach for screening naturally derived ligands for red seabream aryl hydrocarbon receptor

The aryl hydrocarbon receptor (AHR) is a key ligand-dependent transcription factor that mediates the toxic effects of compounds such as dioxin. Recently, natural ligands of AHR, including flavonoids, have been attracting physiological and toxicological attention as they have been reported to regulate major biological functions such as inflammation and anti-cancer by reducing the toxic effects of dioxin. Additionally, it is known that natural AHR ligands can accumulate in wildlife tissues, such as fish. However, studies in fish have investigated only a few ligands in experimental fish species, and the AHR response of marine fish to natural AHR ligands of various other structures has not been thoroughly investigated. To explore various natural AHR ligands in marine fish, which make up the most fish, it is necessary to develop new screening methods that consider the specificity of marine fish. In this study, we investigated the response of natural ligands by constructing in vitro and in silico experimental systems using red seabream as a model species. We attempted to develop a new predictive model to screen potential ligands that can induce transcriptional activation of red seabream AHR1 and AHR2 (rsAHR1 and rsAHR2). This was achieved through multiple analyses using in silico/ in vitro data and Tox21 big data. First, we constructed an in vitro reporter gene assay of rsAHR1 and rsAHR2 and measured the response of 10 representatives natural AHR ligands in COS-7 cells. The results showed that FICZ, Genistein, Daidzein, I3C, DIM, Quercetin and Baicalin induced the transcriptional activity of rsAHR1 and rsAHR2, while Resveratrol and Retinol did not induce the transcriptional activity of rsAHR isoforms. Comparing the EC50 values of the respective compounds in rsAHR1 and rsAHR2, FICZ, Genistein, and Daidzein exhibited similar isoform responses, but I3C, Baicalin, DIM and Quercetin show the isoform-specific responses. These results suggest that natural AHR ligands have specific profiling and transcriptional activity for each rsAHR isoform. In silico analysis, we constructed homology models of the ligand binding domains (LBDs) of rsAHR1 and rsAHR2 and calculated the docking energies (U_dock values) of natural ligands with measured in vitro transcriptional activity and dioxins reported in previous studies. The results showed a significant correlation (R2=0.74(rsAHR1), R2=0.83(rsAHR2)) between docking energy and transcriptional activity (EC50) value, suggesting that the homology model of rsAHR1 and rsAHR2 can be utilized to predict the potential transactivation of ligands. To broaden the applicability of the homology model to diverse compound structures and validate the correlation with transcriptional activity, we conducted additional analyses utilizing Tox21 big data. We calculated the docking energy values for 1860 chemicals in both rsAHR1 and rsAHR2, which were tested for transcriptional activation in Tox21 data against human AHR. By comparing the U_dock energy values between 775 active compounds and 1085 inactive compounds, a significant difference (p<0.001) was observed between the U_dock energy values in the two groups, suggesting that the U_dock value can be applied to distinguish the activation of compounds. Furthermore, we observed a significant correlation (R2=0.45) between the AC50 of Tox21 database and U_dock values of human AHR model. In conclusion, we calculated equations to translate the results of an in silico prediction model for ligand screening of rsAHR1 and rsAHR2 transactivation. This ligand screening model can be a powerful tool to quantitatively estimate AHR transactivation of major marine agents to which red seabream may be exposed. The study introduces a new screening approach for potential natural AHR ligands in marine fish, based on homology model-docking energy values of rsAHR1 and rsAHR2, with implications for future agonist development and applications bridging in silico and in vitro data.

Improving predictions of compound amenability for liquid chromatography-mass spectrometry to enhance non-targeted analysis

Mass-spectrometry-based non-targeted analysis (NTA), in which mass spectrometric signals are assigned chemical identities based on a systematic collation of evidence, is a growing area of interest for toxicological risk assessment. Successful NTA results in better identification of potentially hazardous pollutants within the environment, facilitating the development of targeted analytical strategies to best characterize risks to human and ecological health. A supporting component of the NTA process involves assessing whether suspected chemicals are amenable to the mass spectrometric method, which is necessary in order to assign an observed signal to the chemical structure. Prior work from this group involved the development of a random forest model for predicting the amenability of 5517 unique chemical structures to liquid chromatography-mass spectrometry (LC-MS). This work improves the interpretability of the group's prior model of the same endpoint, as well as integrating 1348 more data points across negative and positive ionization modes. We enhance interpretability by feature engineering, a machine learning practice that reduces the input dimensionality while attempting to preserve performance statistics. We emphasize the importance of interpretable machine learning models within the context of building confidence in NTA identification. The novel data were curated by the labeling of compounds as amenable or unamenable by expert curators, resulting in an enhanced set of chemical compounds to expand the applicability domain of the prior model. The balanced accuracy benchmark of the newly developed model is comparable to performance previously reported (mean CV BA is 0.84 vs. 0.82 in positive mode, and 0.85 vs. 0.82 in negative mode), while on a novel external set, derived from this work's data, the Matthews correlation coefficients (MCC) for the novel models are 0.66 and 0.68 for positive and negative mode, respectively. Our group's prior published models scored MCC of 0.55 and 0.54 on the same external sets. This demonstrates appreciable improvement over the chemical space captured by the expanded dataset. This work forms part of our ongoing efforts to develop models with higher interpretability and higher performance to support NTA efforts.

Adverse effects in traditional and alternative toxicity tests

Chemical safety assessment begins with defining the lowest level of chemical that alters one or more measured endpoints. This critical effect level, along with factors to account for uncertainty, is used to derive limits for human exposure. In the absence of data regarding the specific mechanisms or biological pathways affected, non-specific endpoints such as body weight and non-target organ weight changes are used to set critical effect levels. Specific apical endpoints such as impaired reproductive function or altered neurodevelopment have also been used to set chemical safety limits; however, in test guidelines designed for specific apical effect(s), concurrently measured non-specific endpoints may be equally or more sensitive than specific endpoints. This means that rather than predicting a specific toxicological response, animal data are often used to develop protective critical effect levels, without assuming the same change would be observed in humans. This manuscript is intended to encourage a rethinking of how adverse chemical effects are interpreted: non-specific endpoints from in vivo toxicological studies data are often used to derive points of departure for use with safety assessment factors to create recommended exposure levels that are broadly protective but not necessarily target-specific.