Potential for Machine Learning to Address Data Gaps in Human Toxicity and Ecotoxicity Characterization

Emission Factor Recommendation for Life Cycle Assessments with Generative AI

Accurately quantifying greenhouse gas (GHG) emissions is crucial for organizations to measure and mitigate their environmental impact. Life cycle assessment (LCA) estimates the environmental impacts throughout a product's entire lifecycle, from raw material extraction to end-of-life. Measuring the emissions outside a product owner's control is challenging, and practitioners rely on emission factors (EFs)─estimations of GHG emissions per unit of activity─to model and estimate indirect impacts. However, the current practice of manually selecting appropriate EFs from databases is time-consuming and error-prone and requires expertise. We present an AI-assisted method leveraging natural language processing and machine learning to automatically recommend EFs with human-interpretable justifications. Our algorithm can assist experts by providing a ranked list of EFs or operating in a fully automated manner, where the top recommendation is selected as final. Benchmarks across multiple real-world data sets show our method recommends the correct EF with an average precision of 86.9% in the fully automated case and shows the correct EF in the top 10 recommendations with an average precision of 93.1%. By streamlining EF selection, our approach enables scalable and accurate quantification of GHG emissions, supporting organizations' sustainability initiatives and progress toward net-zero emissions targets across industries.

Quantitative prediction of toxicological points of departure using two-stage machine learning models: A new approach methodology (NAM) for chemical risk assessment

Point of departure (POD) is a concept used in risk assessment to calculate the reference dose of exposure that is likely to have no appreciable risk on health. POD can be directly utilized from no observed adverse effect levels (NOAEL) which is the dose or exposure level at which there is little or no risk of adverse effects. However, NOAEL values are unavailable for most of the chemicals due to inconsistent animal toxicity data. Hence, the current study utilizes a two-stage machine learning (ML) model for predicting NOAEL values, based on data curated from diverse toxicity exposures. In the first stage, a random forest regressor is used for supervised outlier detection and removal addressing any variability in data and poor correlations. The refined data is then used for toxicity prediction using several ML models; random forest and XGBoost show relatively higher performance with an R2 value of 0.4 and 0.43, respectively, for predicting NOAEL in chronic toxicity. Similarly, feature combinations with absorption distribution metabolism and excretion (ADME) indicate better NOAEL prediction for acute toxicity. External validation is performed by predicting NOAEL values for cosmetic pigments and calculating reference doses (RfD). Notably, pigments like orange and red show higher RfD values, indicating broader safety margins. This study provides a practical framework for addressing variability and data limitations in toxicity prediction while offering insights into its applicability in risk evaluation.

Operationalization of the safe and sustainable by design framework for chemicals and materials: challenges and proposed actions

The production and use of chemicals and materials have both advantages and drawbacks for human and ecosystem health. This has led to a demand for carefully guided, safe, and sustainable innovation in the production of chemicals and materials, taking into consideration their entire life cycle. The European Commission's Joint Research Centre (JRC) has released the Safe and Sustainable by Design (SSbD) framework, which aims to support this objective. The SSbD framework consists of two components that are intended to be iteratively implemented throughout the innovation process: (1) the application of design principles phase, and (2) the safety and sustainability assessment phase. However, the operationalization of the framework is currently challenging. This article maps the challenges and proposes ways to address them effectively. The mapping, which is based on a literature review and stakeholder opinions, resulted in 35 challenges. The highest priority challenge is "integration of SSbD framework into the innovation process." To begin addressing this issue, this article recommends conducting a scoping analysis to define the SSbD study. This can be achieved through implementing a tiered approach that aligns with the objectives of the innovation and the growing expertise that comes with it. The second priority challenge is "data availability, quality and uncertainty." This can be supported by using Findability, Accessibility, Interoperability, and Reuse (FAIR) principles and by optimizing in silico methods at early stages of the innovation process. An infrastructure for data and communication is necessary to effectively engage with the entire value chain. The third priority challenge is "integration of safety and sustainability aspects," which requires a clear definition of how to integrate those aspects in the SSbD context, and harmonization, as far as possible, of input data, assumptions, and scenario construction. This review is the first step in accelerating the operationalization of the novel SSbD concept and framework into industrial practice.

AI-aided chronic mixture risk assessment along a small European river reveals multiple sites at risk and pharmaceuticals being the main risk drivers

The vast amount of registered chemicals leads to a high diversity of substances occurring in the environment and the creation of new substances outpaces chemical risk assessment as well as monitoring strategies. Hence, risk assessment strategies need to be modified ensuring that they remain aligned with the rapid development and marketing of new substances. Here we performed a longitudinal chronic mixture risk assessment considering a real-world case study scenario with diverse anthropogenic impact types characterised by different land uses along a river in Central Germany. We sampled river water using large-volume solid phase extraction at six selected sampling sites. Following chemical analysis using liquid chromatography-high resolution mass spectrometry, we quantified 192 substances. For 34 % of them, we obtained empirical chronic effect data for freshwater organisms. Furthermore, we used the open-source artificial intelligence (AI) model TRIDENT to predict chronic toxicity for all substances. A multi-scenario mixture risk assessment was conducted for three taxonomic groups, using the concentration-addition concept and considering various hazard and exposure scenarios. The results showed that the chronic risk estimates for all taxonomic groups were considerably higher when the empirical data was amended with data from in silico modelling. We identified hotspots of chemical pollution and our analysis indicated that fish were the most vulnerable taxonomic group, with pharmaceuticals being the most relevant risk drivers. Our study exemplifies the application of an AI model to predict chronic risk for aquatic organisms in combination with the consideration of multiple risk scenarios that may complement future risk assessment strategies.

Assessing the stereoselective bioactivity and biotoxicity of penthiopyrad in soil environment for efficacy improvement and hazard reduction

Penthiopyrad, a chiral pesticide, has been widely used in agricultural production. However, systematic evaluation of stereoselective bioactivity and biotoxicity of penthiopyrad in soil environment is insufficient. In this study, the stereoselective bioactivity of penthiopyrad against three soil-borne disease pathogens and its stereoselective biotoxicity to soil non-target organisms were investigated. The present results showed that the bioactivities of S-penthiopyrad were 546, 76 and 1.1-fold higher than those of R-penthiopyrad due to their different interaction modes with SDH in different target pathogens. S-penthiopyrad was more persistent in the soil environment and had stronger bioaccumulation than R-penthiopyrad. The accumulation of penthiopyrad in earthworms induced the response of detoxification system, resulting in the significant increases in the activity of detoxifying enzymes, such as GST, CarE, and CYP450. Additionally, both S-penthiopyrad and R-penthiopyrad induced cell apoptosis, intestinal damage and differentially expressed genes in earthworms, especially S-penthiopyrad. Furthermore, S-penthiopyrad has stronger binding capacity with COL6A and ACE proteins, while R-penthiopyrad has stronger binding capacity with CYP450 family proteins, which may be the main reason for the differences in biotoxicity between PEN enantiomers. Considering the differences in bioactivity and biotoxicity of penthiopyrad enantiomers, as well as the modes of action of pesticides on target and non-target organisms, S-penthiopyrad has greater potential for future development.

Two-Stage Machine Learning-Based Approach to Predict Points of Departure for Human Noncancer and Developmental/Reproductive Effects

Chemical points of departure (PODs) for critical health effects are crucial for evaluating and managing human health risks and impacts from exposure. However, PODs are unavailable for most chemicals in commerce due to a lack of in vivo toxicity data. We therefore developed a two-stage machine learning (ML) framework to predict human-equivalent PODs for oral exposure to organic chemicals based on chemical structure. Utilizing ML-based predictions for structural/physical/chemical/toxicological properties from OPERA 2.9 as features (Stage 1), ML models using random forest regression were trained with human-equivalent PODs derived from in vivo data sets for general noncancer effects (n = 1,791) and reproductive/developmental effects (n = 2,228), with robust cross-validation for feature selection and estimating generalization errors (Stage 2). These two-stage models accurately predicted PODs for both effect categories with cross-validation-based root-mean-squared errors less than an order of magnitude. We then applied one or both models to 34,046 chemicals expected to be in the environment, revealing several thousand chemicals of moderate concern and several hundred chemicals of high concern for health effects at estimated median population exposure levels. Further application can expand by orders of magnitude the coverage of organic chemicals that can be evaluated for their human health risks and impacts.

Characterizing Freshwater Ecotoxicity of More Than 9000 Chemicals by Combining Different Levels of Available Measured Test Data with In Silico Predictions

Ecotoxicological impacts of chemicals released into the environment are characterized by combining fate, exposure, and effects. For characterizing effects, species sensitivity distributions (SSDs) estimate toxic pressures of chemicals as the potentially affected fraction of species. Life cycle assessment (LCA) uses SSDs to identify products with lowest ecotoxicological impacts. To reflect ambient concentrations, the Global Life Cycle Impact Assessment Method (GLAM) ecotoxicity task force recently recommended deriving SSDs for LCA based on chronic EC10s (10% effect concentration, for a life-history trait) and using the 20th percentile of an EC10-based SSD as a working point. However, because we lacked measured effect concentrations, impacts of only few chemicals were assessed, underlining data limitations for decision support. The aims of this paper were therefore to derive and validate freshwater SSDs by combining measured effect concentrations with in silico methods. Freshwater effect factors (EFs) and uncertainty estimates for use in GLAM-consistent life cycle impact assessment were then derived by combining three elements: (1) using intraspecies extrapolating effect data to estimate EC10s, (2) using interspecies quantitative structure-activity relationships, or (3) assuming a constant slope of 0.7 to derive SSDs. Species sensitivity distributions, associated EFs, and EF confidence intervals for 9862 chemicals, including data-poor ones, were estimated based on these elements. Intraspecies extrapolations and the fixed slope approach were most often applied. The resulting EFs were consistent with EFs derived from SSD-EC50 models, implying a similar chemical ecotoxicity rank order and method robustness. Our approach is an important step toward considering the potential ecotoxic impacts of chemicals currently neglected in assessment frameworks due to limited test data. Environ Toxicol Chem 2024;43:1914-1927. © 2024 The Author(s). Environmental Toxicology and Chemistry published by Wiley Periodicals LLC on behalf of SETAC.

Probabilistic Reference and 10% Effect Concentrations for Characterizing Inhalation Non-cancer and Developmental/Reproductive Effects for 2,160 Substances

Chemicals assessment and management frameworks rely on regulatory toxicity values, which are based on points of departure (POD) identified following rigorous dose-response assessments. Yet, regulatory PODs and toxicity values for inhalation exposure (i.e., reference concentrations [RfCs]) are available for only ∼200 chemicals. To address this gap, we applied a workflow to determine surrogate inhalation route PODs and corresponding toxicity values, where regulatory assessments are lacking. We curated and selected inhalation in vivo data from the U.S. EPA's ToxValDB and adjusted reported effect values to chronic human equivalent benchmark concentrations (BMCh) following the WHO/IPCS framework. Using ToxValDB chemicals with existing PODs associated with regulatory toxicity values, we found that the 25th %-ile of a chemical's BMCh distribution (POD p 25 BMC h ) could serve as a suitable surrogate for regulatory PODs (Q2 ≥ 0.76, RSE ≤ 0.82 log10 units). We applied this approach to derive POD p 25 BMC h for 2,095 substances with general non-cancer toxicity effects and 638 substances with reproductive/developmental toxicity effects, yielding a total coverage of 2,160 substances. From these POD p 25 BMC h , we derived probabilistic RfCs and human population effect concentrations. With this work, we have expanded the number of chemicals with toxicity values available, thereby enabling a much broader coverage for inhalation risk and impact assessment.

Near-field exposures and human health impacts for organic chemicals in interior paints: A high-throughput screening

Interior paints contain organic chemicals that might be harmful to painters and building residents. This study aims to develop a high-throughput approach to screen near-field human exposures and health impacts related to organic chemicals in interior paints. We developed mass balance models for both water- and solvent-based paints, predicting emissions during wet and dry phases. We then screened exposures and risks, focusing on Sri Lanka where residential houses are frequently repainted. These models accurately predict paint drying time and indoor air concentrations of organic chemicals. Exposures of both painter and household resident were estimated for 65 organic chemicals in water-based and 26 in solvent-based paints, considering 12 solvents. Chemicals of concerns (CoCs) were identified, and maximum acceptable chemical contents (MACs) were calculated. Water-based paints generally pose lower health risks than solvent-based paints but might contain biocides of high concern. The total human health impact of one painting event on all household adults ranges from 1.5 × 10-3 to 2.1 × 10-2 DALYs for solvent-based paints, and from 4.1 × 10-4 to 9.5 × 10-3 DALYs for water-based paints. The present approach is a promising way to support the formulation of safer paint, and is integrated in the USEtox scientific consensus model for use in life cycle assessment, chemical substitution and risk screening.

Quantifying pesticide emissions for drift deposition in comparative risk and impact assessment

Estimating emissions of chemical pesticides used in agriculture is an essential component in evaluating the potential toxicity-related impacts on humans and ecosystems in various comparative risk and impact assessment frameworks, such as life cycle assessment, environmental footprinting, absolute environmental sustainability assessment, chemical substitution, and risk prioritization. Emissions related to drift deposition-usually derived from drift experiments-can reach non-target areas, and vary as a function of crop characteristics and application technique. We derive cumulative drift deposition fractions for a wide range of experimental drift functions for use in comparative and mass-balanced approaches. We clarify that cumulative drift deposition fractions require to integrate the underlying drift functions over the relevant deposition area and to correct for the ratio of deposition area to treated field area to arrive at overall mass deposited per unit mass of applied pesticide. Our results show that for most crops, drift deposition fractions from pesticide application are below 0.03 (i.e. 3% of applied mass), except for grapes and fruit trees, where drift fractions can reach 5% when using canon or air blast sprayers. Notably, aerial applications on soybeans can result in significantly higher drift deposition fractions, ranging from 20% to 60%. Additionally, varying the nozzle position can lead to a factor of five differences in pesticide deposition, and establishing buffer zones can effectively reduce drift deposition. To address remaining limitations in deriving cumulative drift deposition fractions, we discuss possible alternative modelling approaches. Our proposed approach can be implemented in different quantitative and comparative assessment frameworks that require emission estimates of agricultural pesticides, in support of reducing chemical pollution and related impacts on human health and the environment.