Computational Toxicology最新文献_第5页

Towards a quantitative adverse outcome pathway for liver carcinogenesis: From proliferation to prediction 肝癌发生不良后果的定量途径：从增殖到预测

IF 3.1 Q2 TOXICOLOGY

Computational Toxicology

Pub Date : 2025-06-01 DOI: 10.1016/j.comtox.2025.100359

Christina H.J. Veltman , Hiba Khalidi , Elias Zgheib , Bob van de Water , Mirjam Luijten , Jeroen L.A. Pennings

Hazard assessment of non-genotoxic carcinogens could greatly benefit from next generation risk assessment approaches, driven by the multitude of mechanisms through which non-genotoxic carcinogens operate. One method for structuring new approach methodology-derived data is the adverse outcome pathway (AOP) concept. Currently, mostly qualitative AOPs are described, limiting their application for regulatory decision making. In contrast, quantitative AOPs use mathematical terms to describe the relationships between key events (KEs), allowing for the derivation of a Point of Departure (PoD). Here, we report quantification of the key event relationship (KER) between sustained hepatocyte proliferation and liver tumour formation, two KEs of AOP#220 relating to CYP2E1 activation leading to liver cancer. We use incidence of histopathological lesions indicative of proliferation, as well as BrdU labelling obtained from existing sub-chronic toxicity studies in rats, to quantify proliferation. For liver cancer, incidences of hepatocellular adenoma and carcinoma from 2-year rodent carcinogenicity studies were collected. Data for both KEs were combined to calibrate a response-response model, and Bayesian logistic regression analysis was applied to obtain predictions and credible intervals for carcinogenicity. Proliferative lesion incidence was observed to be a highly specific, yet insensitive predictor, and combining this with BrdU labelling yields more accurate predictions of carcinogenicity. Importantly, we demonstrate that for most of the chemicals tested, inclusion of BrdU labelling returns more precise predicted benchmark dose intervals for PoD derivation. To further explore this quantitative KER and its regulatory application, we propose to include and standardize BrdU labelling for sub-chronic toxicity studies performed for regulatory purposes.

非基因毒性致癌物的危害评估可以极大地受益于下一代风险评估方法，这些方法是由非基因毒性致癌物的多种作用机制驱动的。构建新方法方法衍生数据的一种方法是不良结果路径（AOP）概念。目前，对aop的描述大多是定性的，限制了它们在监管决策中的应用。相比之下，定量aop使用数学术语来描述关键事件（ke）之间的关系，从而允许推导出一个起点（PoD）。在这里，我们报告了持续肝细胞增殖和肝脏肿瘤形成之间的关键事件关系（KER）的量化，AOP#220的两个ke与CYP2E1激活导致肝癌有关。我们使用指示增殖的组织病理学病变发生率，以及从现有的大鼠亚慢性毒性研究中获得的BrdU标记来量化增殖。对于肝癌，收集了2年啮齿类动物致癌性研究中肝细胞腺瘤和肝癌的发生率。将两种ke的数据合并以校准响应-响应模型，并应用贝叶斯逻辑回归分析获得致癌性的预测和可信区间。观察到增生性病变发生率是一个高度特异性但不敏感的预测因子，将其与BrdU标记相结合可以更准确地预测致癌性。重要的是，我们证明，对于大多数被测试的化学品，包含BrdU标签可以为PoD衍生提供更精确的预测基准剂量间隔。为了进一步探索定量KER及其监管应用，我们建议将BrdU标签纳入并标准化用于监管目的的亚慢性毒性研究。

{"title":"Towards a quantitative adverse outcome pathway for liver carcinogenesis: From proliferation to prediction","authors":"Christina H.J. Veltman , Hiba Khalidi , Elias Zgheib , Bob van de Water , Mirjam Luijten , Jeroen L.A. Pennings","doi":"10.1016/j.comtox.2025.100359","DOIUrl":"10.1016/j.comtox.2025.100359","url":null,"abstract":"<div><div>Hazard assessment of non-genotoxic carcinogens could greatly benefit from next generation risk assessment approaches, driven by the multitude of mechanisms through which non-genotoxic carcinogens operate. One method for structuring new approach methodology-derived data is the adverse outcome pathway (AOP) concept. Currently, mostly qualitative AOPs are described, limiting their application for regulatory decision making. In contrast, quantitative AOPs use mathematical terms to describe the relationships between key events (KEs), allowing for the derivation of a Point of Departure (PoD). Here, we report quantification of the key event relationship (KER) between sustained hepatocyte proliferation and liver tumour formation, two KEs of AOP#220 relating to CYP2E1 activation leading to liver cancer. We use incidence of histopathological lesions indicative of proliferation, as well as BrdU labelling obtained from existing sub-chronic toxicity studies in rats, to quantify proliferation. For liver cancer, incidences of hepatocellular adenoma and carcinoma from 2-year rodent carcinogenicity studies were collected. Data for both KEs were combined to calibrate a response-response model, and Bayesian logistic regression analysis was applied to obtain predictions and credible intervals for carcinogenicity. Proliferative lesion incidence was observed to be a highly specific, yet insensitive predictor, and combining this with BrdU labelling yields more accurate predictions of carcinogenicity. Importantly, we demonstrate that for most of the chemicals tested, inclusion of BrdU labelling returns more precise predicted benchmark dose intervals for PoD derivation. To further explore this quantitative KER and its regulatory application, we propose to include and standardize BrdU labelling for sub-chronic toxicity studies performed for regulatory purposes.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100359"},"PeriodicalIF":3.1,"publicationDate":"2025-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144253717","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}

引用次数: 0

From blood to body tissues: a dynamic framework for estimating volatile organic compound exposure using Kalman filtering and physiological models 从血液到身体组织：使用卡尔曼滤波和生理模型估计挥发性有机化合物暴露的动态框架

IF 3.1 Q2 TOXICOLOGY

Computational Toxicology

Pub Date : 2025-06-01 DOI: 10.1016/j.comtox.2025.100362

Laurent Simon

Accurate quantification of volatile organic compound (VOC) concentrations in target tissues is critical for robust exposure assessment and toxicological risk analysis. Conventional methods that rely on blood measurements and partition behaviors often fail to consider the transient nature of real-time exposure. Physiologically-based pharmacokinetic (PBPK) models advance predictive capabilities by simulating absorption, distribution, metabolism, and excretion (ADME) processes. However, their accuracy is limited by measurement errors and parameter uncertainties. This study combines the Kalman Filter (KF) with a linear PBPK model (KF-PBPK) to dynamically refine VOC tissue concentration estimates and support real-time exposure assessment using blood measurements. The Kalman Filter is an algorithm that continuously updates model predictions based on new measurements. It filters out noise and improves the accuracy of estimates. The application of the KF-Expectation Maximization (KF-EM) approach to human m-xylene exposure data improved the signal-to-noise ratio (SNR) from 13.9 dB to 17.4 dB. The KF-PBPK scheme effectively captured the multi-compartment kinetics of VOC distribution across several compartments. Filtered estimates closely matched the experimental data, demonstrating the framework’s effectiveness in modeling and predicting human VOC exposure. This research suggests that the KF-PBPK is a reliable tool for improving VOC exposure assessments, with potential implications for environmental pollution monitoring, risk assessment and regulatory decision-making.

准确量化目标组织中的挥发性有机化合物（VOC）浓度对于可靠的暴露评估和毒理学风险分析至关重要。依靠血液测量和分割行为的传统方法往往不能考虑实时暴露的瞬态性质。基于生理的药代动力学（PBPK）模型通过模拟吸收、分布、代谢和排泄（ADME）过程来提高预测能力。然而，它们的精度受到测量误差和参数不确定性的限制。本研究将卡尔曼滤波（KF）与线性PBPK模型（KF-PBPK）相结合，以动态改进VOC组织浓度估计，并支持使用血液测量进行实时暴露评估。卡尔曼滤波是一种基于新测量值不断更新模型预测的算法。它滤除了噪声，提高了估计的准确性。将kf -期望最大化（KF-EM）方法应用于人体间二甲苯暴露数据，将信噪比（SNR）从13.9 dB提高到17.4 dB。KF-PBPK方案有效地捕获了VOC分布的多室动力学。过滤后的估计与实验数据密切匹配，证明了该框架在建模和预测人类VOC暴露方面的有效性。该研究表明，KF-PBPK是改进VOC暴露评估的可靠工具，对环境污染监测、风险评估和监管决策具有潜在意义。

{"title":"From blood to body tissues: a dynamic framework for estimating volatile organic compound exposure using Kalman filtering and physiological models","authors":"Laurent Simon","doi":"10.1016/j.comtox.2025.100362","DOIUrl":"10.1016/j.comtox.2025.100362","url":null,"abstract":"<div><div>Accurate quantification of volatile organic compound (VOC) concentrations in target tissues is critical for robust exposure assessment and toxicological risk analysis. Conventional methods that rely on blood measurements and partition behaviors often fail to consider the transient nature of real-time exposure. Physiologically-based pharmacokinetic (PBPK) models advance predictive capabilities by simulating absorption, distribution, metabolism, and excretion (ADME) processes. However, their accuracy is limited by measurement errors and parameter uncertainties. This study combines the Kalman Filter (KF) with a linear PBPK model (KF-PBPK) to dynamically refine VOC tissue concentration estimates and support real-time exposure assessment using blood measurements. The Kalman Filter is an algorithm that continuously updates model predictions based on new measurements. It filters out noise and improves the accuracy of estimates. The application of the KF-Expectation Maximization (KF-EM) approach to human m-xylene exposure data improved the signal-to-noise ratio (SNR) from 13.9 dB to 17.4 dB. The KF-PBPK scheme effectively captured the multi-compartment kinetics of VOC distribution across several compartments. Filtered estimates closely matched the experimental data, demonstrating the framework’s effectiveness in modeling and predicting human VOC exposure. This research suggests that the KF-PBPK is a reliable tool for improving VOC exposure assessments, with potential implications for environmental pollution monitoring, risk assessment and regulatory decision-making.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100362"},"PeriodicalIF":3.1,"publicationDate":"2025-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144272478","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}

引用次数: 0

Chronic and acute eco-toxicity modeling of carcinogenic and hazardous air pollutants toward humans for critical risk assessment and regulatory decision-making 致癌和有害空气污染物对人类的慢性和急性生态毒性模拟，用于关键风险评估和监管决策

IF 3.1 Q2 TOXICOLOGY

Computational Toxicology

Pub Date : 2025-06-01 DOI: 10.1016/j.comtox.2025.100358

Ankur Kumar , Probir Kumar Ojha , Kunal Roy

Rapid and regular exposure to carcinogenic, toxic, and hazardous chemicals in humans and other living organisms can cause serious chronic (long-term) and acute (short-term) health issues. Since in-vitro and in-vivo toxicity testing requires a long time, a large number of animal experiments, and a high cost, in-silico toxicity testing is the best alternative supported by various regulatory organizations. In our current work, multiple regression-based Quantitative structure–activity relationship models (two chronic toxicity models, a QAAR (quantiative activity-activity relationship) model (chronic studies), and seven acute toxicity models) have been developed to assess the chronic and acute toxicities of carcinogenic chemicals toward humans rigorously following the OECD principles. Statistical validation metrics (R² = 0.604–0.990, Q²_LOO = 0.558––0.988, Q²_F1 = 0.580–0.990, Q²_F2 = 0.503–0.988, MAE_test = 0.103–0.766) demonstrated the robustness, reliability, reproducibility, and predictivity of the developed models. The developed models were utilized to screen the PPDB database, and their predictions were validated against real-world data to confirm their predictive accuracy and reliability. Thus, the present work will significantly aid in bridging the chronic and acute toxicity data gap, identifying carcinogenic chemicals, screening various chemical databases, and developing safer (from observed bio-marker), non-carcinogenic, and greener chemicals strictly obeying the reduction, refinement, and replacement (3Rs) guidelines.

人类和其他生物迅速和经常接触致癌、有毒和危险化学品可导致严重的慢性（长期）和急性（短期）健康问题。由于体外和体内毒性测试需要较长的时间、大量的动物实验和较高的成本，因此硅毒性测试是各监管机构支持的最佳替代方案。在我们目前的工作中，基于多元回归的定量结构-活性关系模型（两个慢性毒性模型，一个定量活性-活性关系模型（慢性研究）和七个急性毒性模型）已经开发出来，严格遵循经合组织的原则来评估致癌化学物质对人类的慢性和急性毒性。统计验证指标（R2 = 0.604 ~ 0.990, Q2LOO = 0.558 ~ 0.988, Q2F1 = 0.580 ~ 0.990, Q2F2 = 0.503 ~ 0.988, MAEtest = 0.103 ~ 0.766）验证了所建立模型的稳健性、可靠性、可重复性和可预测性。开发的模型用于筛选PPDB数据库，并通过实际数据验证其预测的准确性和可靠性。因此，目前的工作将大大有助于弥合慢性和急性毒性数据差距，识别致癌化学物质，筛选各种化学数据库，并开发更安全（从观察到的生物标志物），非致癌和更环保的化学物质，严格遵守减少，改进和替代（3Rs）指南。

{"title":"Chronic and acute eco-toxicity modeling of carcinogenic and hazardous air pollutants toward humans for critical risk assessment and regulatory decision-making","authors":"Ankur Kumar , Probir Kumar Ojha , Kunal Roy","doi":"10.1016/j.comtox.2025.100358","DOIUrl":"10.1016/j.comtox.2025.100358","url":null,"abstract":"<div><div>Rapid and regular exposure to carcinogenic, toxic, and hazardous chemicals in humans and other living organisms can cause serious chronic (long-term) and acute (short-term) health issues. Since <em>in-vitro</em> and <em>in-vivo</em> toxicity testing requires a long time, a large number of animal experiments, and a high cost, in-silico toxicity testing is the best alternative supported by various regulatory organizations. In our current work, multiple regression-based Quantitative structure–activity relationship models (two chronic toxicity models, a QAAR (quantiative activity-activity relationship) model (chronic studies), and seven acute toxicity models) have been developed to assess the chronic and acute toxicities of carcinogenic chemicals toward humans rigorously following the OECD principles. Statistical validation metrics (R<sup>2</sup> = 0.604–0.990, Q<sup>2</sup><sub>LOO</sub> = 0.558––0.988, Q<sup>2</sup><sub>F1</sub> = 0.580–0.990, Q<sup>2</sup><sub>F2</sub> = 0.503–0.988, MAE<sub>test</sub> = 0.103–0.766) demonstrated the robustness, reliability, reproducibility, and predictivity of the developed models. The developed models were utilized to screen the PPDB database, and their predictions were validated against real-world data to confirm their predictive accuracy and reliability. Thus, the present work will significantly aid in bridging the chronic and acute toxicity data gap, identifying carcinogenic chemicals, screening various chemical databases, and developing safer (from observed bio-marker), non-carcinogenic, and greener chemicals strictly obeying the reduction, refinement, and replacement (3Rs) guidelines.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100358"},"PeriodicalIF":3.1,"publicationDate":"2025-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144196203","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}

引用次数: 0

A comparative study of biostatistical pipelines for benchmark concentration modeling of in vitro screening assays 体外筛选试验基准浓度建模的生物统计管道的比较研究

IF 3.1 Q2 TOXICOLOGY

Computational Toxicology

Pub Date : 2025-06-01 DOI: 10.1016/j.comtox.2025.100360

Kelly E. Carstens , Arif Dönmez , Jui-Hua Hsieh , Kristina Bartmann , Katie Paul Friedman , Katharina Koch , Martin Scholze , Ellen Fritsche

New approach methods (NAMs) have been prioritized to reduce the use of animals for chemical safety assessment while continuing to protect human health and the environment. A key challenge of generating toxicity data is the implementation of a standardized analysis approach for transparent and reproducible benchmark concentration (BMC) estimation and uncertainty quantification for assay developers, regulators, and other stakeholders. In this study, we compared the bioactivity results of 321 chemical samples from four established BMC analysis pipelines used for evaluation of developmental neurotoxicity (DNT) NAMs data: the ToxCast pipeline (tcpl), CRStats, DNT DIVER (Curvep and Hill pipelines). We found an overall activity hit call concordance of 77.2 % and highly correlated BMC estimations (r = 0.92 ± 0.02 SD), demonstrating generally good agreement across pipelines. Discordance appeared to be explained predominantly by noise within the data and borderline activity (activity occuring near the benchmark response level). Evaluation of the BMC confidence intervals indicated that pipeline selection may impact the estimation of the BMC lower bound. Consideration of biphasic models appeared important for capturing biologically-relevant changes in activity in the DNT battery. Lastly, different approaches to compute ‘selective’ bioactivity (activity below the threshold of cytotoxicity) were compared, identifying the CRstats classification model as more stringent for classifying selective activity. Overall, these findings indicated greater confidence in NAMs bioactivity results and emphasize the importance of understanding strengths and uncertainties of concentration–response modeling pipelines for informing biological interpretation and application decision making.

新的方法（NAMs）已得到优先考虑，以减少使用动物进行化学品安全评估，同时继续保护人类健康和环境。生成毒性数据的一个关键挑战是为检测开发人员、监管机构和其他利益相关者实施透明和可重复的基准浓度（BMC）估计和不确定度量化的标准化分析方法。在这项研究中，我们比较了用于评估发育神经毒性（DNT） NAMs数据的四种已建立的BMC分析管道中的321种化学样品的生物活性结果：ToxCast管道（tcpl）， CRStats管道，DNT DIVER （curve和Hill管道）。我们发现，总体活动的呼叫一致性为77.2%，BMC估计高度相关（r = 0.92±0.02 SD），表明管道之间的一致性总体良好。不一致似乎主要由数据中的噪声和边界活动（在基准响应水平附近发生的活动）来解释。对BMC置信区间的评估表明，管道选择可能会影响BMC下界的估计。考虑双相模型对于捕获DNT电池活性的生物学相关变化似乎很重要。最后，对计算“选择性”生物活性（低于细胞毒性阈值的活性）的不同方法进行了比较，确定CRstats分类模型对于分类选择性活性更为严格。总的来说，这些发现表明了对NAMs生物活性结果的更大信心，并强调了理解浓度-反应建模管道的优势和不确定性对于为生物学解释和应用决策提供信息的重要性。

{"title":"A comparative study of biostatistical pipelines for benchmark concentration modeling of in vitro screening assays","authors":"Kelly E. Carstens , Arif Dönmez , Jui-Hua Hsieh , Kristina Bartmann , Katie Paul Friedman , Katharina Koch , Martin Scholze , Ellen Fritsche","doi":"10.1016/j.comtox.2025.100360","DOIUrl":"10.1016/j.comtox.2025.100360","url":null,"abstract":"<div><div>New approach methods (NAMs) have been prioritized to reduce the use of animals for chemical safety assessment while continuing to protect human health and the environment. A key challenge of generating toxicity data is the implementation of a standardized analysis approach for transparent and reproducible benchmark concentration (BMC) estimation and uncertainty quantification for assay developers, regulators, and other stakeholders. In this study, we compared the bioactivity results of 321 chemical samples from four established BMC analysis pipelines used for evaluation of developmental neurotoxicity (DNT) NAMs data: the ToxCast pipeline (tcpl), CRStats, DNT DIVER (Curvep and Hill pipelines). We found an overall activity hit call concordance of 77.2 % and highly correlated BMC estimations (r = 0.92 ± 0.02 SD), demonstrating generally good agreement across pipelines. Discordance appeared to be explained predominantly by noise within the data and borderline activity (activity occuring near the benchmark response level). Evaluation of the BMC confidence intervals indicated that pipeline selection may impact the estimation of the BMC lower bound. Consideration of biphasic models appeared important for capturing biologically-relevant changes in activity in the DNT battery. Lastly, different approaches to compute ‘selective’ bioactivity (activity below the threshold of cytotoxicity) were compared, identifying the CRstats classification model as more stringent for classifying selective activity. Overall, these findings indicated greater confidence in NAMs bioactivity results and emphasize the importance of understanding strengths and uncertainties of concentration–response modeling pipelines for informing biological interpretation and application decision making.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100360"},"PeriodicalIF":3.1,"publicationDate":"2025-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144240741","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}

引用次数: 0

The first report on chronic toxicity assessment of metals towards Ceriodaphnia dubia using QSTR technique: A step towards healthier and safer human health and eco-system 利用QSTR技术评价金属对斑点斑切蚤的慢性毒性：迈向更健康、更安全的人类健康和生态系统的一步

IF 3.1 Q2 TOXICOLOGY

Computational Toxicology

Pub Date : 2025-05-27 DOI: 10.1016/j.comtox.2025.100357

Ankur Kumar , Joyita Roy , Probir Kumar Ojha

Exposure of humans and other living organisms to metals (including heavy metals) can lead to serious chronic and acute health effects, which may sometimes be life-threatening. As a result, assessing the toxicity of heavy metals is essential. However, experimental toxicity data for heavy metals is limited, and their toxicity estimation can be highly costly, lengthy analysis durations, and may require animal testing. Therefore, in-silico approaches such as quantitative structure–activity relationship (QSAR) are a suitable alternative. In this work, we have developed multi-endpoints MLR-QSAR models to assess the chronic toxicity of heavy metal towards Ceriodaphnia dubia using 48 data points and obeying the Organization for Economic Cooperation and Development (OECD) guidelines. Intra-endpoint uni-variate models were developed to fill the toxicity data gaps between the endpoints (acute to chronic). The statistical results of the developed models (individual models M1-M4; R² = 0.691–0.738, Q²_LOO = 0.542–0.578, Q²_F1 = 0.673–0.732, Q²_F2 = 0.552–0.580, MAE_95%data = 0.437–0.753; intra-endpoints models IEM1-IEM9; R² = 0.952–0.988, Q²_LOO = 0.907–0.987, Q²_F1 = 0.885–0.991, Q²_F2 = 0.979–0.991, MAE_95%data = 0.120–0.436) infer that the models are robust, reliable, reproducible, and predictive. The descriptors contributing to the development of the model imply that the release of electrons, formation of cations, higher electronegativity, and the presence of neutrons in the heavy metals significantly influence the toxicity caused by the metals. Thus, this study presents in silico models aimed at controlling the exposure of living organisms to toxic heavy metals. It assesses both acute and chronic toxicity, addresses gaps in toxicity data, and strives to create healthier and safer ecosystems by strictly following the principles of reduction, replacement, and refinement (the RRR framework).

人类和其他生物接触金属（包括重金属）可导致严重的慢性和急性健康影响，有时可能危及生命。因此，评估重金属的毒性至关重要。然而，重金属的实验毒性数据是有限的，其毒性估计可能非常昂贵，分析持续时间长，并且可能需要动物试验。因此，像定量构效关系（QSAR）这样的计算机方法是一个合适的选择。在这项工作中，我们开发了多端点MLR-QSAR模型，使用48个数据点并遵循经济合作与发展组织（OECD）的指导方针来评估重金属对dubia Ceriodaphnia的慢性毒性。建立了终点内单变量模型，以填补终点（急性到慢性）之间的毒性数据空白。已开发模型的统计结果(单个模型M1-M4；R2 = 0.691 - -0.738, Q2LOO = 0.542 - -0.578, Q2F1 = 0.673 - -0.732, Q2F2 = 0.552 - -0.580, MAE95%data = 0.437 - -0.753;终端内模型IEM1-IEM9；R2 = 0.952 ~ 0.988, Q2LOO = 0.907 ~ 0.987, Q2F1 = 0.885 ~ 0.991, Q2F2 = 0.979 ~ 0.991, MAE95%data = 0.120 ~ 0.436)表明模型稳健、可靠、可重复性好。有助于模型发展的描述符表明，重金属中电子的释放、阳离子的形成、较高的电负性和中子的存在显著地影响了金属引起的毒性。因此，本研究提出了旨在控制生物体暴露于有毒重金属的硅模型。它评估急性和慢性毒性，填补毒性数据的空白，并通过严格遵循减少、替代和改进（RRR框架）的原则，努力创造更健康、更安全的生态系统。

{"title":"The first report on chronic toxicity assessment of metals towards Ceriodaphnia dubia using QSTR technique: A step towards healthier and safer human health and eco-system","authors":"Ankur Kumar , Joyita Roy , Probir Kumar Ojha","doi":"10.1016/j.comtox.2025.100357","DOIUrl":"10.1016/j.comtox.2025.100357","url":null,"abstract":"<div><div>Exposure of humans and other living organisms to metals (including heavy metals) can lead to serious chronic and acute health effects, which may sometimes be life-threatening. As a result, assessing the toxicity of heavy metals is essential. However, experimental toxicity data for heavy metals is limited, and their toxicity estimation can be highly costly, lengthy analysis durations, and may require animal testing. Therefore, <em>in-silico</em> approaches such as quantitative structure–activity relationship (QSAR) are a suitable alternative. In this work, we have developed multi-endpoints MLR-QSAR models to assess the chronic toxicity of heavy metal towards <em>Ceriodaphnia dubia</em> using 48 data points and obeying the Organization for Economic Cooperation and Development (OECD) guidelines. Intra-endpoint uni-variate models were developed to fill the toxicity data gaps between the endpoints (acute to chronic). The statistical results of the developed models (individual models M1-M4; R<sup>2</sup> = 0.691–0.738, Q<sup>2</sup><sub>LOO</sub> = 0.542–0.578, Q<sup>2</sup><sub>F1</sub> = 0.673–0.732, Q<sup>2</sup><sub>F2</sub> = 0.552–0.580, MAE<sub>95%data</sub> = 0.437–0.753; intra-endpoints models IEM1-IEM9; R<sup>2</sup> = 0.952–0.988, Q<sup>2</sup><sub>LOO</sub> = 0.907–0.987, Q<sup>2</sup><sub>F1</sub> = 0.885–0.991, Q<sup>2</sup><sub>F2</sub> = 0.979–0.991, MAE<sub>95%data</sub> = 0.120–0.436) infer that the models are robust, reliable, reproducible, and predictive. The descriptors contributing to the development of the model imply that the release of electrons, formation of cations, higher electronegativity, and the presence of neutrons in the heavy metals significantly influence the toxicity caused by the metals. Thus, this study presents <em>in silico</em> models aimed at controlling the exposure of living organisms to toxic heavy metals. It assesses both acute and chronic toxicity, addresses gaps in toxicity data, and strives to create healthier and safer ecosystems by strictly following the principles of reduction, replacement, and refinement (the RRR framework).</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100357"},"PeriodicalIF":3.1,"publicationDate":"2025-05-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144168106","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}

引用次数: 0

Prediction of progesterone receptor binding potency, agonism and antagonism using machine learning models 使用机器学习模型预测黄体酮受体结合效力、激动作用和拮抗作用

IF 3.1 Q2 TOXICOLOGY

Computational Toxicology

Pub Date : 2025-05-11 DOI: 10.1016/j.comtox.2025.100351

Nemanja Milošević , Nataša Sukur Milošević , Svetlana Fa Nedeljkovic , Bojana Stanic , Nebojsa Andric

The use of Machine Learning (ML) models to predict the binding potency of chemicals to estrogen and androgen receptors has become well-established, helping in the prioritization of chemicals for endocrine disruption testing. However, the potential of ML models for other endocrine targets, such as the progesterone receptor (PR), remains underexplored. In this study, we developed an ML model to predict PR binding affinity and assess the agonistic/antagonistic properties of chemicals. The model achieved a training accuracy of 99.72% and a validation accuracy of 74.46%. External validation was conducted on a dataset of approximately 10,000 chemicals, including 5720 compounds from the training set for which there is a known outcome. External predictions aligned closely with experimental in vitro data, achieving an accuracy of 96.85%. Additionally, the model successfully predicted PR binding affinity and agonistic/antagonistic properties for chemicals without available experimental data. In summary, this study highlights the potential of ML as an effective tool for prioritizing chemicals for future in vitro and in vivo testing of PR binding potency and agonistic/antagonistic properties of chemicals.

使用机器学习（ML）模型来预测化学物质与雌激素和雄激素受体的结合能力已经得到了很好的应用，这有助于在内分泌干扰测试中确定化学物质的优先级。然而，ML模型对其他内分泌靶点（如孕激素受体（PR））的潜力仍未得到充分探索。在这项研究中，我们建立了一个ML模型来预测PR的结合亲和力，并评估化学物质的激动/拮抗特性。该模型的训练准确率为99.72%，验证准确率为74.46%。外部验证是在大约10,000种化学物质的数据集上进行的，其中包括来自已知结果的训练集的5720种化合物。外部预测与体外实验数据密切吻合，准确率为96.85%。此外，该模型在没有实验数据的情况下成功预测了PR的结合亲和力和化学物质的激动/拮抗特性。总之，本研究强调了ML作为一种有效工具的潜力，可以在体外和体内测试化学物质的PR结合效力和激动/拮抗特性，从而优先考虑化学物质。

{"title":"Prediction of progesterone receptor binding potency, agonism and antagonism using machine learning models","authors":"Nemanja Milošević , Nataša Sukur Milošević , Svetlana Fa Nedeljkovic , Bojana Stanic , Nebojsa Andric","doi":"10.1016/j.comtox.2025.100351","DOIUrl":"10.1016/j.comtox.2025.100351","url":null,"abstract":"<div><div>The use of Machine Learning (ML) models to predict the binding potency of chemicals to estrogen and androgen receptors has become well-established, helping in the prioritization of chemicals for endocrine disruption testing. However, the potential of ML models for other endocrine targets, such as the progesterone receptor (PR), remains underexplored. In this study, we developed an ML model to predict PR binding affinity and assess the agonistic/antagonistic properties of chemicals. The model achieved a training accuracy of 99.72% and a validation accuracy of 74.46%. External validation was conducted on a dataset of approximately 10,000 chemicals, including 5720 compounds from the training set for which there is a known outcome. External predictions aligned closely with experimental <em>in vitro</em> data, achieving an accuracy of 96.85%. Additionally, the model successfully predicted PR binding affinity and agonistic/antagonistic properties for chemicals without available experimental data. In summary, this study highlights the potential of ML as an effective tool for prioritizing chemicals for future <em>in vitro</em> and <em>in vivo</em> testing of PR binding potency and agonistic/antagonistic properties of chemicals.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100351"},"PeriodicalIF":3.1,"publicationDate":"2025-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144068772","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}

引用次数: 0

An R-based predictive model for skin-sensitizing potential of substances with known structures 一种基于r的已知结构物质致敏电位预测模型

IF 3.1 Q2 TOXICOLOGY

Computational Toxicology

Pub Date : 2025-05-11 DOI: 10.1016/j.comtox.2025.100350

Yuri Hatakeyama , Kosuke Imai , Hayato Nishida , Shiho Oeda , Tomomi Atobe , Morihiko Hirota

Evaluation of skin-sensitizing potential is important to confirm the safety of cosmetics. As animal testing is no longer permitted, several alternative methods based on the adverse outcome pathway (AOP) approach have been reported. In addition, integrated approaches to testing and assessment (IATA), which combine the results of multiple alternative methods to assess skin sensitization potential, have been developed. We have reported an artificial neural network (ANN) model for sensitization risk assessment using commercial software, QwikNet. In the present study, we constructed a new sensitization prediction model for substances with known structures using the free and open-source software R for statistical analysis, and compared the results with those of the QwikNet model. The R model was confirmed to show similar predictive performance for estimated concentration three (EC3) which is the concentration of a test substance needed to produce a stimulation index of 3 to the QwikNet model on the same training set of 134 compounds. The accuracy, overpredicted rate, and underpredicted rate of the R model were 81.3%, 10.4%, and 8.2%, respectively, versus 79.9%, 10.4%, and 9.7% for the QwikNet model. In case studies of compounds not included in the training set, the R model showed generally good predictive ability. For less-well-predicted substances, additional in silico and read-across evaluations complemented the ANN model and improved the predictive accuracy. This study demonstrates that the ANN model is portable to the R software system. Furthermore, the combination of ANN prediction with in silico predictions and read-across taking account of substructures improves the prediction of skin-sensitizing potential in a weight-of-evidence approach.

皮肤致敏电位的评估对于确认化妆品的安全性非常重要。由于动物试验不再被允许，一些基于不良结果通路（AOP）方法的替代方法已经被报道。此外，已经开发了综合测试和评估方法（IATA），将多种替代方法的结果结合起来评估皮肤致敏潜力。我们报道了一个人工神经网络（ANN）模型，用于使用商业软件QwikNet进行敏化风险评估。在本研究中，我们利用免费开源软件R对已知结构物质构建了新的敏化预测模型进行统计分析，并与QwikNet模型进行了比较。在134个化合物的同一训练集上，R模型对产生刺激指数为3所需的测试物质的浓度（EC3）的预测性能与QwikNet模型相似。R模型的准确率、高估率和低估率分别为81.3%、10.4%和8.2%，而QwikNet模型的准确率为79.9%、10.4%和9.7%。在未包含在训练集中的化合物的案例研究中，R模型显示出良好的预测能力。对于预测较差的物质，额外的计算机和读取评估补充了人工神经网络模型，提高了预测准确性。研究表明，该人工神经网络模型可移植到R软件系统。此外，将人工神经网络预测与计算机预测和考虑子结构的读取相结合，在证据权重方法中提高了对皮肤致敏电位的预测。

{"title":"An R-based predictive model for skin-sensitizing potential of substances with known structures","authors":"Yuri Hatakeyama , Kosuke Imai , Hayato Nishida , Shiho Oeda , Tomomi Atobe , Morihiko Hirota","doi":"10.1016/j.comtox.2025.100350","DOIUrl":"10.1016/j.comtox.2025.100350","url":null,"abstract":"<div><div>Evaluation of skin-sensitizing potential is important to confirm the safety of cosmetics. As animal testing is no longer permitted, several alternative methods based on the adverse outcome pathway (AOP) approach have been reported. In addition, integrated approaches to testing and assessment (IATA), which combine the results of multiple alternative methods to assess skin sensitization potential, have been developed. We have reported an artificial neural network (ANN) model for sensitization risk assessment using commercial software, QwikNet. In the present study, we constructed a new sensitization prediction model for substances with known structures using the free and open-source software R for statistical analysis, and compared the results with those of the QwikNet model. The R model was confirmed to show similar predictive performance for estimated concentration three (EC3) which is the concentration of a test substance needed to produce a stimulation index of 3 to the QwikNet model on the same training set of 134 compounds. The accuracy, overpredicted rate, and underpredicted rate of the R model were 81.3%, 10.4%, and 8.2%, respectively, versus 79.9%, 10.4%, and 9.7% for the QwikNet model. In case studies of compounds not included in the training set, the R model showed generally good predictive ability. For less-well-predicted substances, additional <em>in silico</em> and read-across evaluations complemented the ANN model and improved the predictive accuracy. This study demonstrates that the ANN model is portable to the R software system. Furthermore, the combination of ANN prediction with <em>in silico</em> predictions and read-across taking account of substructures improves the prediction of skin-sensitizing potential in a weight-of-evidence approach.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100350"},"PeriodicalIF":3.1,"publicationDate":"2025-05-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144089788","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}

引用次数: 0

Evaluating marginal likelihood approximations of dose–response relationship models in Bayesian benchmark dose methods for risk assessment 评估风险评估中贝叶斯基准剂量方法中剂量-反应关系模型的边际似然近似

IF 3.1 Q2 TOXICOLOGY

Computational Toxicology

Pub Date : 2025-05-10 DOI: 10.1016/j.comtox.2025.100347

Sota Minewaki , Tomohiro Ohigashi , Takashi Sozu

Benchmark dose (BMD; a dose associated with a specified change in response) is used to determine the point of departure for the acceptable daily intake of substances for humans. Multiple dose–response relationship models are considered in the BMD method. The Bayesian model averaging (BMA) method is commonly used, where several models are averaged based on their posterior probabilities, which are determined by calculating the marginal likelihood (ML). Several ML approximation methods are employed in standard software packages, such as BBMD, ToxicR, and the EFSA platform for the BMD method, because the ML cannot be analytically calculated. Although ML values differ among approximation methods, resulting in BMD estimates, this phenomenon is neither widely recognized nor quantitatively evaluated. In this study, we evaluated the agreement of BMD estimates among five ML approximation methods in the BMA method. The five ML approximation methods are (1) maximum likelihood estimation (MLE)-based Schwarz criterion, (2) Markov chain Monte Carlo (MCMC)-based Schwarz criterion, (3) Laplace approximation, (4) density estimation, and (5) bridge sampling. We used eight dose–response relationship models and three prior distributions used in BBMD and ToxicR for 518 experimental datasets. The agreement among the approximation methods tended to be low in the non-informative prior distribution. Although the agreements tended to be high in the informative prior distribution, they were low in some approximation methods. Since the approximation method and the prior distribution affect the agreement, their selection should be carefully considered when implementing BMD methods.

基准剂量；与特定反应变化有关的剂量用于确定人体每日可接受物质摄入量的起始点。BMD方法考虑了多种剂量-反应关系模型。贝叶斯模型平均（BMA）是常用的方法，其中几个模型的平均是基于它们的后验概率，这是通过计算边际似然（ML）确定的。由于ML不能解析计算，因此在标准软件包中采用了几种ML近似方法，例如BBMD， ToxicR和EFSA平台的BMD方法。虽然各种近似方法的ML值不同，导致BMD估计，但这种现象既没有得到广泛认识，也没有得到定量评估。在这项研究中，我们评估了bmma方法中五种ML近似方法的BMD估计的一致性。这五种机器学习近似方法分别是：(1)基于最大似然估计（MLE）的Schwarz准则，(2)基于马尔可夫链蒙特卡罗（MCMC）的Schwarz准则，(3)拉普拉斯近似，(4)密度估计，(5)桥式抽样。我们对518个实验数据集使用了BBMD和ToxicR中使用的8个剂量-反应关系模型和3个先验分布。在非信息先验分布中，近似方法之间的一致性往往较低。虽然一致性在信息先验分布中趋于高，但在某些近似方法中却较低。由于近似方法和先验分布会影响一致性，因此在实现BMD方法时应仔细考虑它们的选择。

{"title":"Evaluating marginal likelihood approximations of dose–response relationship models in Bayesian benchmark dose methods for risk assessment","authors":"Sota Minewaki , Tomohiro Ohigashi , Takashi Sozu","doi":"10.1016/j.comtox.2025.100347","DOIUrl":"10.1016/j.comtox.2025.100347","url":null,"abstract":"<div><div>Benchmark dose (BMD; a dose associated with a specified change in response) is used to determine the point of departure for the acceptable daily intake of substances for humans. Multiple dose–response relationship models are considered in the BMD method. The Bayesian model averaging (BMA) method is commonly used, where several models are averaged based on their posterior probabilities, which are determined by calculating the marginal likelihood (ML). Several ML approximation methods are employed in standard software packages, such as BBMD, <span>ToxicR</span>, and the EFSA platform for the BMD method, because the ML cannot be analytically calculated. Although ML values differ among approximation methods, resulting in BMD estimates, this phenomenon is neither widely recognized nor quantitatively evaluated. In this study, we evaluated the agreement of BMD estimates among five ML approximation methods in the BMA method. The five ML approximation methods are (1) maximum likelihood estimation (MLE)-based Schwarz criterion, (2) Markov chain Monte Carlo (MCMC)-based Schwarz criterion, (3) Laplace approximation, (4) density estimation, and (5) bridge sampling. We used eight dose–response relationship models and three prior distributions used in BBMD and <span>ToxicR</span> for 518 experimental datasets. The agreement among the approximation methods tended to be low in the non-informative prior distribution. Although the agreements tended to be high in the informative prior distribution, they were low in some approximation methods. Since the approximation method and the prior distribution affect the agreement, their selection should be carefully considered when implementing BMD methods.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100347"},"PeriodicalIF":3.1,"publicationDate":"2025-05-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143947516","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}

引用次数: 0

Validation of OECD QSAR Toolbox profilers for genotoxicity assessment of pesticides using the MultiCase genotoxicity database 使用多酶遗传毒性数据库进行农药遗传毒性评估的OECD QSAR工具箱分析器的验证

IF 3.1 Q2 TOXICOLOGY

Computational Toxicology

Pub Date : 2025-05-10 DOI: 10.1016/j.comtox.2025.100356

Monika Kemény , Colin M. North , Felix M. Kluxen , Markus Frericks , Dragana Vukelic , Sishuo Cao , Roustem Saiakhov , Mounika Girireddy

Quantitative Structure Activity Relationship (QSAR) models are widely used for genotoxicity assessment in regulatory settings. In silico profilers are a special case of models capturing mechanistic insights specific to a particular toxicological endpoint or reflecting chemistry-related attributes that may not be directly associated with a defined mechanism of toxicity. This study explores the accuracy of using such profilers as a lower tier in genotoxicity assessment to inform regulatory concerns. Relevant profilers in the OECD QSAR Toolbox are investigated using an external validation dataset derived from the MultiCASE Genotoxicity database, which contains AMES mutagenicity and in vivo micronucleus (MNT) experimental results. The MNT dataset includes the commercial in vivo MNT dataset expanded with pesticide data from regulatory documents. This analysis incorporates the use of metabolism simulations by the OECD QSAR Toolbox to assess their influence on profiler performance. The present findings show that the absence of profiler alerts correlates well with experimentally negative outcomes. However, the calculated accuracy for the MNT-related and AMES-related profilers varies considerably (41%-78% for MNT-related profilers and 62%-88% for AMES-related profilers using the full set with and without consideration of metabolism). Incorporating metabolism simulations increases accuracy by 4–6% for the full AMES-dataset, and 4–16% for the full MNT-dataset. Together, genotoxicity assessment using the Toolbox profilers should include a critical evaluation of any triggered alerts, considering the overall performance statistics of the profilers presented within this work. Results from third-party QSAR models provide critical insights to complement the expert review of any profiler positive result, as profilers alone are not recommended to be used directly for prediction purpose.

定量构效关系（QSAR）模型被广泛用于遗传毒性评估。计算机分析器是一种特殊的模型，它捕获特定于特定毒理学终点的机制见解，或反映可能与定义的毒性机制不直接相关的化学相关属性。本研究探讨了在遗传毒性评估中使用这种分析器作为较低层次的准确性，以告知监管方面的关注。使用来自MultiCASE基因毒性数据库的外部验证数据集（其中包含AMES诱变性和体内微核（MNT）实验结果）对OECD QSAR工具箱中的相关分析进行了研究。MNT数据集包括商业体内MNT数据集，扩展了来自监管文件的农药数据。该分析结合了经合组织QSAR工具箱代谢模拟的使用，以评估其对分析器性能的影响。目前的研究结果表明，缺乏分析警报与实验阴性结果密切相关。然而，mnt相关的和ames相关的分析器的计算精度差别很大（mnt相关的分析器的计算精度为41%-78%，ames相关的分析器的计算精度为62%-88%）。结合代谢模拟可使完整mes数据集的准确性提高4-6%，完整mnt数据集的准确性提高4-16%。总之，考虑到本工作中提供的分析程序的总体性能统计数据，使用工具箱分析程序的遗传毒性评估应包括对任何触发警报的关键评估。来自第三方QSAR模型的结果提供了重要的见解，以补充任何分析器阳性结果的专家审查，因为不建议单独使用分析器直接用于预测目的。

{"title":"Validation of OECD QSAR Toolbox profilers for genotoxicity assessment of pesticides using the MultiCase genotoxicity database","authors":"Monika Kemény , Colin M. North , Felix M. Kluxen , Markus Frericks , Dragana Vukelic , Sishuo Cao , Roustem Saiakhov , Mounika Girireddy","doi":"10.1016/j.comtox.2025.100356","DOIUrl":"10.1016/j.comtox.2025.100356","url":null,"abstract":"<div><div>Quantitative Structure Activity Relationship (QSAR) models are widely used for genotoxicity assessment in regulatory settings. <em>In silico</em> profilers are a special case of models capturing mechanistic insights specific to a particular toxicological endpoint or reflecting chemistry-related attributes that may not be directly associated with a defined mechanism of toxicity. This study explores the accuracy of using such profilers as a lower tier in genotoxicity assessment to inform regulatory concerns. Relevant profilers in the OECD QSAR Toolbox are investigated using an external validation dataset derived from the MultiCASE Genotoxicity database, which contains AMES mutagenicity and in vivo micronucleus (MNT) experimental results. The MNT dataset includes the commercial in vivo MNT dataset expanded with pesticide data from regulatory documents. This analysis incorporates the use of metabolism simulations by the OECD QSAR Toolbox to assess their influence on profiler performance. The present findings show that the absence of profiler alerts correlates well with experimentally negative outcomes. However, the calculated accuracy for the MNT-related and AMES-related profilers varies considerably (41%-78% for MNT-related profilers and 62%-88% for AMES-related profilers using the full set with and without consideration of metabolism). Incorporating metabolism simulations increases accuracy by 4–6% for the full AMES-dataset, and 4–16% for the full MNT-dataset. Together, genotoxicity assessment using the Toolbox profilers should include a critical evaluation of any triggered alerts, considering the overall performance statistics of the profilers presented within this work. Results from third-party QSAR models provide critical insights to complement the expert review of any profiler positive result, as profilers alone are not recommended to be used directly for prediction purpose.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100356"},"PeriodicalIF":3.1,"publicationDate":"2025-05-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144089757","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}

引用次数: 0

IF 3.1 Q2 TOXICOLOGY

Computational Toxicology

Pub Date : 2025-05-08 DOI: 10.1016/j.comtox.2025.100353

Brett Hagan , Imran Shah , Grace Patlewicz

Read-across is a technique used to fill data gaps for substances lacking specific hazard data. The technique relies on identifying source analogues with relevant data that are ‘similar’ to the substance of interest (target). Typically, source analogues are identified on the basis of structural similarity but the evaluation of their suitability for read-across depends on other contexts of similarity. This manuscript aimed to review the ways in which source analogues are identified for read-across using chemical fingerprint/scaffold approaches before describing graph-based approaches including; graph kernel, graph embedding, and deep learning. To demonstrate how these could be practically used for analogue identification, five different toxicity datasets of varying size and diversity were selected that had been the subject of previous read-across or QSAR analyses. One dataset was an analogue set whereas the other four datasets comprised substances evaluated for their skin sensitisation, skin irritation, fathead minnow aquatic toxicity and genotoxicity potential. The analogues and their associated similarities using the different graph based approaches were compared with the outcomes from two chemical fingerprint approaches (ToxPrints and Morgan). The results for each dataset are briefly described. Based on the examples evaluated, graph kernel approaches were found to have some promise, in contrast unsupervised whole graph embedding approaches were ineffective for all the datasets evaluated. Graph convolutional networks produced meaningful embeddings for the genotoxicity dataset evaluated. Depending on use case, availability and size of training data, graph similarity approaches have the potential to play a larger role in analogue identification and evaluation for read-across.

跨读是一种技术，用于填补缺乏具体危害数据的物质的数据空白。该技术依赖于识别具有与感兴趣的物质（目标）“相似”的相关数据的源类似物。通常，源相似物是基于结构相似性来识别的，但对其是否适合跨读的评估取决于其他相似上下文。本文旨在回顾在描述基于图的方法之前，使用化学指纹/支架方法识别源类似物的方法，包括；图核，图嵌入，深度学习。为了演示这些如何实际用于类似物鉴定，选择了五个不同大小和多样性的不同毒性数据集，这些数据集已成为先前读取或QSAR分析的主题。一个数据集是模拟集，而其他四个数据集包括评估其皮肤致敏，皮肤刺激，黑头鲦鱼水生毒性和遗传毒性潜力的物质。使用不同的基于图的方法得到的相似物及其相关的相似性与两种化学指纹方法（ToxPrints和Morgan）的结果进行了比较。简要描述了每个数据集的结果。基于评估的示例，发现图核方法有一定的前景，相比之下，无监督全图嵌入方法对所有评估的数据集都无效。图卷积网络为评估的遗传毒性数据集产生了有意义的嵌入。根据用例、可用性和训练数据的大小，图相似方法有可能在跨读的模拟识别和评估中发挥更大的作用。

{"title":"Can graph similarity metrics be helpful for analogue identification as part of a read-across approach?","authors":"Brett Hagan , Imran Shah , Grace Patlewicz","doi":"10.1016/j.comtox.2025.100353","DOIUrl":"10.1016/j.comtox.2025.100353","url":null,"abstract":"<div><div>Read-across is a technique used to fill data gaps for substances lacking specific hazard data. The technique relies on identifying source analogues with relevant data that are ‘similar’ to the substance of interest (target). Typically, source analogues are identified on the basis of structural similarity but the evaluation of their suitability for read-across depends on other contexts of similarity. This manuscript aimed to review the ways in which source analogues are identified for read-across using chemical fingerprint/scaffold approaches before describing graph-based approaches including; graph kernel, graph embedding, and deep learning. To demonstrate how these could be practically used for analogue identification, five different toxicity datasets of varying size and diversity were selected that had been the subject of previous read-across or QSAR analyses. One dataset was an analogue set whereas the other four datasets comprised substances evaluated for their skin sensitisation, skin irritation, fathead minnow aquatic toxicity and genotoxicity potential. The analogues and their associated similarities using the different graph based approaches were compared with the outcomes from two chemical fingerprint approaches (ToxPrints and Morgan). The results for each dataset are briefly described. Based on the examples evaluated, graph kernel approaches were found to have some promise, in contrast unsupervised whole graph embedding approaches were ineffective for all the datasets evaluated. Graph convolutional networks produced meaningful embeddings for the genotoxicity dataset evaluated. Depending on use case, availability and size of training data, graph similarity approaches have the potential to play a larger role in analogue identification and evaluation for read-across.</div></div>","PeriodicalId":37651,"journal":{"name":"Computational Toxicology","volume":"34 ","pages":"Article 100353"},"PeriodicalIF":3.1,"publicationDate":"2025-05-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143935958","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}

引用次数: 0