Skin sensitization is a significant concern for chemical safety assessments. Traditional animal assays often fail to predict human responses accurately, and ethical constraints limit the collection of human data, necessitating a need for reliable in silico models of skin sensitization prediction. This study introduces HuSSPred, an in silico tool based on the Human Predictive Patch Test (HPPT). HuSSPred aims to enhance the reliability of predicting human skin sensitization effects for chemical agents to support their regulatory assessment. We have curated an extensive HPPT database and performed chemical space analysis and grouping. Binary and multiclass QSAR models were developed with Bayesian hyperparameter optimization. Model performance was evaluated via five-fold cross-validation. We performed model validation with reference data from the Defined Approaches for Skin Sensitization (DASS) app. HuSSPred models demonstrated strong predictive performance with CCR ranging from 55 to 88%, sensitivity between 48 and 89%, and specificity between 37 and 92%. The positive predictive value (PPV) ranged from 84 to 97%, versus negative predictive value (NPV) from 22 to 65%, and coverage was between 75 and 93%. Our models exhibited comparable or improved performance compared to existing tools, and the external validation showed the high accuracy and sensitivity of the developed models. HuSSPred provides a reliable, open-access, and ethical alternative to traditional testing for skin sensitization. Its high accuracy and reasonable coverage make it a valuable resource for regulatory assessments, aligning with the 3Rs principles. The publicly accessible HuSSPred web tool offers a user-friendly interface for predicting skin sensitization based on chemical structure.
Background: Understanding potential prenatal and development toxicity hazard associated with the use of pharmaceutical and cosmetic products is an important component of women health. This hazard can be estimated from chemical structure of respective agents using Quantitative Structure-Activity Relationship (QSAR) models; however, the development of reliable models is challenging due to the complex nature of this endpoint. Methods: Aggregating and curating data from the Food and Drug Administration (FDA), Teratogen Information System (TERIS) database, and select independent studies, we have created, to the best of our knowledge, the largest publicly available dataset comprising compounds annotated as developmental toxicants or not toxicants. Results: We built several binary classification QSAR models exhibiting a correct classification rate of 62-72%, a sensitivity of 66-75%, a specificity of 59-82%, and high coverage of 70-90% assessed using five-fold external validation protocol. We developed a publicly accessible web portal PregPred for developmental toxicity prediction of both overall toxicity and trimester-specific toxicity predictions. Conclusions: Due to high accuracy and coverage as well as public accessibility of the respective web portal, our models can be employed as a computational tool to support regulatory assessment of pharmaceutical and cosmetic products in alignment with the 3Rs (refining, reducing, and replacing) of animal testing. This in silico model holds the potential to substantially influence the field of developmental toxicology, steering regulatory practices toward safer drug development for pregnant women. The first-of-its-kind curated dataset of developmental toxicants and all developed models implemented as a user-friendly web tool, PregPred, are freely available at https://pregpred.mml.unc.edu/).
Artificial intelligence (AI) and machine learning are transforming toxicological research and chemical safety assessment. Although user-friendly computational toxicology platforms are increasingly available, integrating, customizing, and deploying AI methods within end-to-end workflows still often requires programming expertise. This barrier increases the time to adoption of new methods and slows regulatory uptake. To address this limitation, we survey recent initiatives democratizing computational toxicology through no-code/low-code pipelines, automated workflows, and open-source tools. We emphasize solutions for four computational needs: (i) data extraction and access, (ii) data mining and curation, (iii) data analysis and visualization, and (iv) modeling and prediction. These initiatives transform complex computational methods into guided and web-accessible applications that enable toxicologists, regulators, and researchers to leverage AI without coding expertise. The broad applicability of computational methods will be essential for supporting and scaling federal initiatives that advance human-relevant alternatives to animal testing. We also offer practical considerations for domain-specific tool development, including large language model-based information extraction, chemical structure standardization, interactive chemical grouping, and the development of validated machine learning models, as used in the Modeling and Visualization (MoVIZ) pipeline. The authors map the future of computational toxicology and cheminformatics, one that does not require scientists to become programmers but rather makes sophisticated AI tools more broadly accessible, transparent, and guided through thoughtful interface design, transparent workflows, and open science initiatives.
Over the past several decades, reducing, refining, and replacing animal testing (three R’s) has been a prominent goal in chemical toxicology.1 The STopTox (Systemic and Topical chemical Toxicity) platform was developed for this objective as an innovative in-silico alternative to conventional animal testing for acute systemic and topical toxicity testing.2 STopTox utilizes quantitative structure-activity relationship (QSAR) models to predict the toxicity of chemicals, providing a comprehensive, accessible, and user-friendly tool for hazard identification.2 STopTox models were rigorously validated during its initial development using extensive publicly available data sets, ensuring compliance with the Organisation for Economic Co-operation and Development (OECD) principles. These models boasted high internal accuracy and substantial external predictive power.2,3 Despite these promising results, continued validation with novel compounds is integral in establishing the robustness and reliability needed for STopTox to be used as a substitute for in vivo animal testing. In this research letter, we aim to evaluate the predictive performance of STopTox using independent data sets across the six major endpoints of acute toxicity: acute oral, dermal, and inhalation systemic toxicity, as well as skin sensitization, skin irritation/corrosion, and eye irritation/corrosion through external validation. The outcomes of this validation underscore the potential of STopTox to reliably predict toxicity, thereby supporting STopTox as a reliable regulatory decision-making tool that contributes to reducing animal testing in toxicological assessments.