{"title":"An exploration of the use of hybrid fingerprints in Generalized Read-Across and their impact on predictive performance for selected in vivo toxicity outcomes","authors":"Aubrey Leary , Imran Shah , Grace Patlewicz","doi":"10.1016/j.comtox.2025.100349","DOIUrl":"10.1016/j.comtox.2025.100349","url":null,"abstract":"<div><div>Read-across is a cost-efficient means of generating information for hazard assessment. Approaches such as Generalized Read-Across (GenRA) facilitate objective and reproducible read-across for untested substances. GenRA is a web application, and its prediction engine is also available as a python package (genra-py). Recent updates permit source analogues to be identified using ‘hybrid’ fingerprints, i.e. analogues identified based on more than one type of similarity measure. Herein, the performance of hybrid fingerprints relative to Morgan chemical fingerprints was evaluated for a selection of acute and chronic <em>in vivo</em> toxicity outcomes. Grid search and cross-validation on a dataset of 5,830 chemicals with rodent acute oral toxicity (LD<sub>50</sub>) values were used to tune the hybrid weight hyperparameter for up to four chemical fingerprints (Morgan, Torsion, ToxPrint and Analog Identification Methodology (AIM)). The optimal hybrid fingerprint derived (52.12% Morgan, 23.40% ToxPrint, 12.44% AIM, 12.04% Torsion) outperformed Morgan fingerprints across all 10 folds of a cross-validation procedure (mean test set coefficient of determination (R<sup>2</sup>) 0.517 (Morgan) vs. 0.557 (hybrid)). The hybrid fingerprint was then used to make toxicity predictions for 2 other datasets, a set of 3,266 chemicals with oral chronic human equivalent benchmark dose values (mean test set R<sup>2</sup> 0.445 vs. 0.417 for Morgan) and a set of 9,443 chemicals with acute mammalian oral hazard classifications (mean balanced accuracy (BA) 0.577 vs 0.553 for Morgan). Overall, performance improved when using the hybrid fingerprint tuned for the acute toxicity dataset. Using the custom hybrid option in GenRA results in improved read-across predictions relative to current defaults.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100349"},"PeriodicalIF":3.1,"publicationDate":"2025-04-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143887946","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Ryoichi Murakami , Mika Imamura , Masakazu Tateshita , Hajime Kojima , Yasushi Hikida
{"title":"Consensus model for skin sensitization assessment using a rule-based model and LLNA and GPMT statistics-based models","authors":"Ryoichi Murakami , Mika Imamura , Masakazu Tateshita , Hajime Kojima , Yasushi Hikida","doi":"10.1016/j.comtox.2025.100348","DOIUrl":"10.1016/j.comtox.2025.100348","url":null,"abstract":"<div><div>The potential for skin sensitization has traditionally been assessed <em>in vivo</em>; however, animal welfare concerns, the trend toward restrictions, and the prohibition of the use of animals have led to a shift toward the use of non-animal alternatives such as <em>in vitro</em> and <em>in silico</em> tools. <em>In silico</em> tools mainly include rule-based and statistics-based models. Although the use of multiple computational methods is recommended, many tools consist of only one method. Furthermore, skin sensitization develops through multiple key event (KE)/adverse outcome (AO) pathways, but many <em>in silico</em> tools consist of only one KE/AO. We constructed a consensus model based on three different independent skin sensitization KE/AOs from a rule-based model, a local lymph node assay (LLNA) statistics-based model, and a guinea pig maximization test (GPMT) statistics-based model. The rule-based model is based on KE1 and considers the metabolism of pre- and pro-haptens. The LLNA and GPMT statistics-based models are based on KE4 and AO, respectively, and characterized by the use of approximately 2000 and 3000 chemicals in the training dataset, respectively. These models use larger datasets than those previously reported. The constructed consensus model was tested on chemicals labeled with human results from OECD Guideline 497. The results showed that the performance of the majority-voting model was the highest, with a balanced accuracy of 78%. The model combines a wide range of chemical spaces with high prediction accuracy.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100348"},"PeriodicalIF":3.1,"publicationDate":"2025-04-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143874794","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Optimal experimental designs for big and small experiments in toxicology with applications to studying hormesis via metaheuristics","authors":"Brian P.H. Wu , Ray-Bing Chen , Weng Kee Wong","doi":"10.1016/j.comtox.2025.100345","DOIUrl":"10.1016/j.comtox.2025.100345","url":null,"abstract":"<div><div>There are theoretical methods for constructing model-based optimal designs for a given design criterion when the sample size is large. Some of these methods may work for certain models or design criteria and some may find the optimal designs only under a restrictive setting. When the sample size is small, the theory-based methods may become invalid and the optimal designs may also not be implementable. Our first goal is to introduce nature-inspired metaheuristics to efficiently find all types of model-based optimal designs. These metaheuristic algorithms, widely used in engineering, computer science, and artificial intelligence, are generally fast and free of stringent assumptions. For our second goal, we introduce an efficient rounding method to produce an implementable, exact design for small-sized experiments based on large-sample optimal designs. To provide toxicologists with easy access to a variety of model-based optimal designs for both large and small experiments, our third goal is to develop a web-based app. This app will generate different types of model-based optimal designs, allow comparisons, and evaluate the efficiency of any design. As an application, we focus on hormesis and find model-based designs for detecting the presence of hormesis, estimating model parameters and estimating the threshold dose. The methodology is not restricted to studying hormesis only and is broadly applicable for designing other studies in toxicology and beyond.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100345"},"PeriodicalIF":3.1,"publicationDate":"2025-04-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143834957","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Maryam , Mobeen Ur Rehman , Kil to Chong , Hilal Tayara
{"title":"Drugs inhibition prediction in P-gp enzyme: a comparative study of machine learning and graph neural network","authors":"Maryam , Mobeen Ur Rehman , Kil to Chong , Hilal Tayara","doi":"10.1016/j.comtox.2025.100344","DOIUrl":"10.1016/j.comtox.2025.100344","url":null,"abstract":"<div><div>Drug metabolism is a complex and highly regulated process that involves the safe breakdown and elimination of drugs from the body through chemical reactions. The P-glycoprotein (P-gp) plays a key role in drug metabolism, and interfere of drugs with its transport function leads to drug toxicity. Therefore, predicting P-gp inhibition is crucial to avoid adverse drug effects. To address this, machine learning and deep learning models offer a powerful approach to accurately predict the P-gp inhibition. In this study, we have utilized a publicly available P-gp dataset to develop classification models using various machine learning algorithms (SVM, RFC, HistGradient Boosting, AdaBoost) and graph neural networks. The dataset was transformed into molecular descriptors and graph feature vectors to explore the chemical space of metabolic enzymes. Our experimental results demonstrate that machine learning models outperform deep learning models in terms of accuracy and efficiency for independent datasets. Among all models, SVM exhibited superior predictive capabilities for the P-gp data set with an accuracy of 0.95 on independent datasets. Furthermore, the analysis of the importance of the characteristics of the best model highlighted the significant contributions of specific descriptors to the data set. Finally, our model outperformed previous studies when evaluated on an external dataset, emphasizing the efficacy of molecular features in providing more precise explanations of compound properties and biological activity.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100344"},"PeriodicalIF":3.1,"publicationDate":"2025-04-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143808553","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"Benchmark doses (BMD) extrapolated from in vitro cytotoxicity experiments in SH-SY5Y cells using the EFSA Bayesian BMD web app: The study case of imidacloprid","authors":"Lenin J. Ramirez-Cando , Santiago J. Ballaz","doi":"10.1016/j.comtox.2025.100346","DOIUrl":"10.1016/j.comtox.2025.100346","url":null,"abstract":"<div><div>Identifying possible chemical hazards is critical to establish toxicological risk assessment related to non-cancer health effects. The benchmark dose (BMD) is an estimate of the hazardous toxic level (dose or concentration) that produces a predetermined variation in the response rate of an adverse effect (exposure-related risk endpoint). Our goal was to assess how well Bayesian weighted averaging models enhances the imidacloprid-treated SH-SY5Y cells’ dose-toxicological response in the MTT cytotoxicity assay. Notably, the Gelman-Rubin statistics for our models were constantly between 0.9 and 1.0, and the effective sample size was greater than 150, which guaranteed practical sufficiency. We used a weighted average of posteriorly fitted models to estimate the final BMD = 26.40 and the lower confidence limit (BMDL) = 13.10. Including uncertainty factors (UF) in conjunction with MTT data into our risk analysis, we assessed the population’s imidacloprid exposure. The Point of Departure (PoD) at 5th percentil (8.14) indicated adverse effects. Moreover, a similar link was observed between the target human dose for minimal impact (HDMI) and the HD<sub>50</sub> (dose hazardous to 50 % of people). The determined Reference Dose (RfD) of 0.0003 µM suggested a high toxicity risk associated with imidacloprid exposure. Summarizing, dose–response evaluations were enhanced by Bayesian model averaging, highlighting the significance of probabilistic modeling and toxicological understanding.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100346"},"PeriodicalIF":3.1,"publicationDate":"2025-04-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143776894","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
{"title":"A multicomponent similarity approach to identify potential substances of very high concern","authors":"Yordan Yordanov , Emiel Rorije , Jordi Minnema , Thimo Schotman , Willie J.G.M. Peijnenburg , Pim N.H. Wassenaar","doi":"10.1016/j.comtox.2025.100343","DOIUrl":"10.1016/j.comtox.2025.100343","url":null,"abstract":"<div><div>The number of chemicals being placed on the market is increasing. As such, there is an increased need for screening and evaluation of chemical hazards and risks. Particularly, chemicals with intrinsic properties that are considered of very high concern are ideally identified and regulated before wide-spread use and exposure. The use of <em>in silico</em> tools can help to identify potential substances of very high concern (SVHCs).</div><div>Earlier, predictive models have been developed that identify potential SVHCs based on global structural similarity to known SVHCs. Here in this study, these read-across similarity models have been extended with other similarity modules, including toxicophore, biological and physicochemical similarity.</div><div>The newly developed SVHC similarity profiles do individually not outperform the existing global similarity model. However, combining these new modules in an extended similarity approach results in more comprehensive predictions and allows for improved interpretability and applicability to the broader chemical universe. As such, this new approach is thought to support model users in interpretation of the model-prediction, and can thereby contribute to better screening and prioritization of potential SVHCs.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100343"},"PeriodicalIF":3.1,"publicationDate":"2025-04-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143792157","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Matthias M. Wehr , Hilda Witters , Silvie Remy , Bruce Schultz , Marc Jacobs , Sylvia E. Escher
{"title":"Feasibility study on integrating morphological and transcriptomic data for identifying teratogenic molecular markers in zebrafish embryo toxicity testing","authors":"Matthias M. Wehr , Hilda Witters , Silvie Remy , Bruce Schultz , Marc Jacobs , Sylvia E. Escher","doi":"10.1016/j.comtox.2025.100342","DOIUrl":"10.1016/j.comtox.2025.100342","url":null,"abstract":"<div><div>To create a resource for the integration of developmental toxicity new approach methodologies (NAM) data, zebrafish embryo toxicity test (ZET) data were curated and the zeTera database was created. To capture observations of the morphological alteration potential of chemicals, the zeTera database contains experimental study designs and morphological observation data from the literature. Observations of alterations recorded in zeTera were mapped to ontologies and terms were harmonized. In addition, public transcriptomics repositories were mined for data on zebrafish embryos under chemically induced stress. The re-analyzed datasets were compiled into the zetOmics database for the identification of biomarkers of teratogenicity. To identify data-rich compounds, an overlap of both databases was formed, and compounds were grouped based on structural similarities.</div><div>To identify the molecular drivers of teratogenic toxicants, Triadimefon was chosen as model compound for its well-documented teratogenic effects and known mode of action (MOA). We have compiled existing data about Triadimefon from zeTera and conducted additional testing using the ZET, with additional gene expression measurements for data gap filling. From the literature search we identified the adverse outcome pathway (AOP) of triadimefon leading to craniofacial malformations by disruption of retinoic acid metabolism.</div><div>Transcriptomic response in a concentration dependent manner was observed as early as 24 h post fertilization (hpf) with consistent, statistically significant, differential expression spanning the later timepoints. A set of 5 genes (<em>cyp26a1</em>, <em>dhrs3b</em>, <em>cyp26b1</em>, <em>cthrc1a</em>, and <em>cd248b</em>) were selected for their differential expression pattern across time and concentration. These biomarkers were further confirmed using read across approach including data from related structures.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100342"},"PeriodicalIF":3.1,"publicationDate":"2025-03-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143738669","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
David Inauen , Leonie Sophie Lautz , Aalbert Jan Hendriks , Ronette Gehring
{"title":"Augmented allometric scaling: Predicting drug clearance in farm animals with machine learning using body weight","authors":"David Inauen , Leonie Sophie Lautz , Aalbert Jan Hendriks , Ronette Gehring","doi":"10.1016/j.comtox.2025.100341","DOIUrl":"10.1016/j.comtox.2025.100341","url":null,"abstract":"<div><div>In farm animals, kinetic data of exogenous chemicals, such as pharmaceuticals, environmental pollutants or feed contaminants, are scarce. To allow extrapolation across chemicals and species this study developed a machine learning approach that integrated allometric scaling and quantitative structure–activity relationships to predict total body clearance in farm animals. Using body weight and molecular descriptors of chemicals, the models applied both linear and non-linear machine learning methods such as random forest to predict clearance. Data for intravenously administered chemicals were collected from literature from a variety of species. Molecular descriptors of these chemicals were computed. Log-transformed clearances were predicted for five farm animal species—cattle, sheep, goat, swine, horse—as well as dogs and cats for comparative analysis. Two models using machine learning methods were developed: a purely extrapolative machine learning model, and an approach titled “augmented allometric scaling” which, similarly to simple allometric scaling, used pre-existing data in other species to predict a chemicals’ clearance in a target species. The extrapolative approach had large differences in training and test set metrics, while the latter approach demonstrated modestly improved predictive accuracy over simple allometric scaling in farm animals with up to 60.8% of predictions below a fold error of 2, compared to 51% given by allometry, with a difference of up to 0.5 fold errors. In dogs, the new approach performed comparably and worse in cats. This study highlights potentials and limits of machine learning in refining kinetic predictions in farm animals.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"33 ","pages":"Article 100341"},"PeriodicalIF":3.1,"publicationDate":"2025-03-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143580294","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Henk J. van Lingen , Edoardo Saccenti , Maria Suarez-Diez , Marta Baccaro , Nico W. van den Brink
{"title":"Predicting uptake and elimination kinetics of chemicals in invertebrates: A technical note on residual variance modeling","authors":"Henk J. van Lingen , Edoardo Saccenti , Maria Suarez-Diez , Marta Baccaro , Nico W. van den Brink","doi":"10.1016/j.comtox.2024.100337","DOIUrl":"10.1016/j.comtox.2024.100337","url":null,"abstract":"<div><div>Toxicokinetic models for predicting contents of nanomaterials and other toxic chemicals are often fitted without evaluation of the residual variance structure. The aim of the present study was to evaluate various residual variance structures, assuming either homoscedasticity or heteroscedasticity, when fitting non-linear toxicokinetic one-compartment models for predicting uptake, bioaccumulation and elimination of chemicals in invertebrate organisms. Data describing the exposure of several aquatic and terrestrial invertebrates to specific metal nanomaterials and other chemicals were available from real experiments for evaluating the residual variance functions for toxicokinetic models. As proof of concept, datasets of truly homoscedastic and heteroscedastic nature were simulated. Depending the dataset, applying models with different residuals variance assumption largely affected the residual plots and the error margins of parameters or the predicted content of a chemical. Consequently, selecting the most accurate residual variance functions for toxicokinetic modeling, either homoscedastic or heteroscedastic, improves the prediction of chemical contents in invertebrate organisms and the estimation of the associated uptake and elimination rates.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"33 ","pages":"Article 100337"},"PeriodicalIF":3.1,"publicationDate":"2024-12-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143147585","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Upendra K. Devisetty , Emma De Neef , Eric R.L. Gordon , Valeria Velásquez-Zapata , Kenneth Narva , Laurent Mézin , Peter Mc Cahon , Kenneth W. Witwer , Krishnakumar Sridharan
{"title":"A bioinformatics framework for human health risk assessment of externally applied dsRNA-based biopesticides","authors":"Upendra K. Devisetty , Emma De Neef , Eric R.L. Gordon , Valeria Velásquez-Zapata , Kenneth Narva , Laurent Mézin , Peter Mc Cahon , Kenneth W. Witwer , Krishnakumar Sridharan","doi":"10.1016/j.comtox.2024.100340","DOIUrl":"10.1016/j.comtox.2024.100340","url":null,"abstract":"<div><div>Current plant protection methods rely predominantly on conventional chemical pesticides that can have negative human health and environmental impacts. Consequently, there is a pressing need to develop sustainable crop protection solutions that have improved safety profiles for humans and other non-target organisms (NTOs). RNA interference (RNAi) is a natural defense mechanism against viruses found in eukaryotes that silences viral genes in a sequence-specific manner. Recently, RNAi has been utilized to specifically target essential genes of pests with a novel class of topical, sprayable biopesticides based on dsRNA (double-stranded RNA). A critical step in the regulatory approval of such externally applied dsRNA-based biopesticides is a robust bioinformatics analysis of potential off-target effects to humans and other organisms. However, no generally applicable guidelines are available for risk assessment of dsRNA-based biopesticides for humans. Here, we address this gap by describing a bioinformatics framework for risk assessment in humans, informed by peer-reviewed literature, that quantifies potential off-targets with a primary focus on externally applied dsRNA-based biopesticides. The framework comprises three main components: bioinformatics tools for predicting off-target effects in humans, a mismatch tolerance for sequence divergence between dsRNA and unintended targets to delineate potential human off-target effects, and siRNA criteria for quantifying the possibility of theoretical gene silencing in the presence of mismatches in humans. This bioinformatics framework represents the most comprehensive approach described to date and has been used successfully for evaluating the potential risks of the externally applied dsRNA-based biopesticide Calantha<sup>TM</sup> to humans.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"33 ","pages":"Article 100340"},"PeriodicalIF":3.1,"publicationDate":"2024-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143147584","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}