The role of trust in the use of artificial intelligence for chemical risk assessment.

Risk assessment of chemicals is a time-consuming process and needs to be optimized to ensure all chemicals are timely evaluated and regulated. This transition could be stimulated by valuable applications of in silico Artificial Intelligence (AI)/Machine Learning (ML) models. However, implementation of AI/ML models in risk assessment is lagging behind. Most AI/ML models are considered 'black boxes' that lack mechanistical explainability, causing risk assessors to have insufficient trust in their predictions. Here, we explore 'trust' as an essential factor towards regulatory acceptance of AI/ML models. We provide an overview of the elements of trust, including technical and beyond-technical aspects, and highlight elements that are considered most important to build trust by risk assessors. The results provide recommendations for risk assessors and computational modelers for future development of AI/ML models, including: 1) Keep models simple and interpretable; 2) Offer transparency in the data and data curation; 3) Clearly define and communicate the scope/intended purpose; 4) Define adoption criteria; 5) Make models accessible and user-friendly; 6) Demonstrate the added value in practical settings; and 7) Engage in interdisciplinary settings. These recommendations should ideally be acknowledged in future developments to stimulate trust and acceptance of AI/ML models for regulatory purposes.

ZZS similarity tool: The online tool for similarity screening to identify chemicals of potential concern.

Screening and prioritization of chemicals is essential to ensure that available evaluation capacity is invested in those substances that are of highest concern. We, therefore, recently developed structural similarity models that evaluate the structural similarity of substances with unknown properties to known Substances of Very High Concern (SVHC), which could be an indication of comparable effects. In the current study the performance of these models is improved by (1) separating known SVHCs in more specific subgroups, (2) (re-)optimizing similarity models for the various SVHC-subgroups, and (3) improving interpretability of the predicted outcomes by providing a confidence score. The improvements are directly incorporated in a freely accessible web-based tool, named the ZZS similarity tool: https://rvszoeksysteem.rivm.nl/ZzsSimilarityTool. Accordingly, this tool can be used by risk assessors, academia and industrial partners to screen and prioritize chemicals for further action and evaluation within varying frameworks, and could support the identification of tomorrow's substances of concern.

Characterization of ecotoxicological risks from unintentional mixture exposures calculated from European freshwater monitoring data: Forwarding prospective chemical risk management.

Current regulatory chemical safety assessments do not acknowledge that ambient exposures are to multiple chemicals at the same time. As a result, potentially harmful exposures to unintentional mixtures may occur, leading to potential insufficient protection of the environment. The present study describes cumulative environmental risk assessment results for European fresh water ecosystems, based on the NORMAN chemical surface water monitoring database (1998-2016). It aims to characterize the magnitude of the mixture problem and the relative contribution of chemicals to the mixture risk, and evaluates how cumulative risks reduce when the acceptable risk per single chemical is fractionally lowered. Available monitoring data were curated and aggregated to 26,631 place-time combinations with at least two chemicals, of which 376 place-time combinations had at least 25 chemicals identified above the Limit of Detection. Various risk metrics were based on measured environmental concentrations (MECs). Mixture risk characterization ratio's (ΣRCRs) ≥ 1 were found for 39% of the place-time combinations, with few chemicals dominating the ΣRCR. Analyses of mixture toxic pressures, expressed as multi-substance Potentially Affected Fractions of species based on No Observed Effect Concentrations (msPAFNOEC), showed similar outcomes. Small fractional reductions of the ambient chemical concentrations give a steep increase of the percentage of sufficiently protected water bodies (i.e. ΣRCR < 1 and msPAFNOEC < 5%). Scientific and regulatory aspects of these results are discussed, especially with reference to the representativeness of the monitoring data for characterizing ambient mixtures, the robustness of the findings, and the possible regulatory implementation of the concept of a Mixture Allocation Factor (MAF) for prospective chemicals risk management. Although the monitoring data do not represent the full spectrum of ambient mixture exposures in Europe, results show the need for adapting policies to reach European Union goals for a toxic-free environment and underpin the utility and possible magnitude of a MAF.

Persistence, bioaccumulation and toxicity-assessment of petroleum UVCBs: A case study on alkylated three-ring PAHs.

Substances with (very) persistent, (very) bioaccumulative, and/or toxic properties (PBT/vPvB) are of environmental concern and are identified via hazard-based PBT-assessment approaches. The PBT-assessment of well-defined substances is optimized over the past decades, but is under development for substances of unknown or variable composition, complex reaction products or biological materials (UVCBs). Particularly, the large number of constituents and variable composition complicate the PBT-assessment of UVCBs. For petroleum UVCBs, the use of the hydrocarbon block method (HBM) is proposed. Within this method, groups of constituents with similar physicochemical properties and structure are treated as a single entity and are expected to have comparable environmental fate and hazard properties. So far, however, there is a lack of experience with the application of the HBM for PBT-assessment purposes. The aim of this study is to investigate the suitability of the HBM for the PBT-assessment of petroleum UVCBs by evaluating the group of alkylated three-ring polycyclic aromatic hydrocarbons (PAHs). The presented approach is based on experimental data and model predictions and followed the guidelines of the European Chemicals Agency. Because of a lack of relevant experimental data, relative trend analyses were applied. The results indicate that alkylated three-ring PAHs are more persistent, bioaccumulative, and toxic than the parent three-ring PAHs. As the parent three-ring PAHs are currently identified within Europe as PBT/vPvB substances, the alkylated three-ring PAHs could also be considered as PBT/vPvB. Accordingly, this case study provides the prospects for the application of the HBM for the PBT-assessment of UVCBs using trend analysis.

Evaluating chemical similarity as a measure to identify potential substances of very high concern.

Due to the large amount of chemical substances on the market, fast and reproducible screening is essential to prioritize chemicals for further evaluation according to highest concern. We here evaluate the performance of structural similarity models that are developed to identify potential substances of very high concern (SVHC) based on structural similarity to known SVHCs. These models were developed following a systematic analysis of the performance of 112 different similarity measures for varying SVHC-subgroups. The final models consist of the best combinations of fingerprint, similarity coefficient and similarity threshold, and suggested a high predictive performance (≥80%) on an internal dataset consisting of SVHC and non-SVHC substances. However, the application performance on an external dataset was not evaluated. Here, we evaluated the application performance of the developed similarity models with a 'pseudo-external assessment' on a set of substances (n = 60-100 for the varying SVHC-subgroups) that were putatively assessed as SVHC or non-SVHC based upon consensus scoring using expert elicitations (n = 30 experts). Expert scores were direct evaluations based on structural similarity to the most similar SVHCs according to the similarity models, and did not consider an extensive evaluation of available data. The use of expert opinions is particularly suitable as this is exactly the intended purpose of the chemical similarity models: a quick, reproducible and automated screening tool that mimics the expert judgement that is frequently applied in various screening applications. In addition, model predictions were analyzed via qualitative approaches and discussed via specific examples, to identify the model's strengths and limitations. The results indicate a good statistical performance for carcinogenic, mutagenic or reprotoxic (CMR) and endocrine disrupting (ED) substances, whereas a moderate performance was observed for (very) persistent, (very) bioaccumulative and toxic (PBT/vPvB) substances when compared to expert opinions. For the PBT/vPvB model, particularly false positive substances were identified, indicating the necessity of outcome interpretation. The developed similarity models are made available as a freely-accessible online tool. In general, the structural similarity models showed great potential for screening and prioritization purposes. The models proved to be effective in identifying groups of substances of potential concern, and could be used to identify follow-up directions for substances of potential concern.

Variability in fish bioconcentration factors: Influences of study design and consequences for regulation.

The fish bioconcentration factor (BCF) is an important aspect within bioaccumulation assessments. Several factors have been suggested to influence BCF values - including species, developmental stage, mixture exposure, and calculation method. However, their exact contribution to variance in BCF values is unknown. Within this study we assessed the relative impact of these test characteristics on BCF values and analyzed the reproducibility of aquatic exposure bioconcentration tests. Linear mixed effects analyses were performed on a newly develop database to investigate the relationship between the response variable (i.e. lipid normalized log BCF values) and several test characteristics as fixed effects. Lower BCF values were observed for substances that were simultaneously applied with high molecular weight polycyclic aromatic hydrocarbons compared to single substance exposure (with an average difference of -0.81 log BCF). Also, lower BCFs upon kinetic determination were observed compared to steady-state BCFs (log BCF -0.27), and lower BCFs for species from the Ostariophysi subcohort level (log BCF -0.17 to -0.15). In addition, data analysis showed high variation within BCF values for single substances (average SD = log BCF 0.21), which questions the robustness of the current bioaccumulation assessments. For example, the 95% confidence range of a BCF value of 2500 ranges from 953 ('not-bioaccumulative') to 6561 ('very bioaccumulative'). Our results show that the use of one single BCF leads to a high uncertainty in bioaccumulation assessments. We strongly recommend that within future bioconcentration studies, the used experimental design and test conditions are described in detail and justified to support solid interpretation.

Pesticide Mixture Toxicity in Surface Water Extracts in Snails (Lymnaea stagnalis) by an in Vitro Acetylcholinesterase Inhibition Assay and Metabolomics.

Many chemicals in use end up in the aquatic environment. The toxicity of water samples can be tested with bioassays, but a metabolomic approach has the advantage that multiple end points can be measured simultaneously and the affected metabolic pathways can be revealed. A current challenge in metabolomics is the study of mixture effects. This study aims at investigating the toxicity of an environmental extract and its most abundant chemicals identified by target chemical analysis of >100 organic micropollutants and effect-directed analysis (EDA) using the acetylcholinesterase (AChE) bioassay and metabolomics. Surface water from an agricultural area was sampled with a large volume solid phase extraction (LVSPE) device using three cartridges containing neutral, anionic, and cationic sorbents able to trap several pollutants classes like pharmaceuticals, pesticides, PAHs, PCBs, and perfluorinated surfactants. Targeted chemical analysis and AChE bioassay were performed on the cartridge extracts. The extract of the neutral sorbent cartridge contained most of the targeted chemicals, mainly imidacloprid, thiacloprid, and pirimicarb, and was the most potent AChE inhibitor. Using an EDA approach, other AChE inhibiting candidates were identified in the neutral extract, such as carbendazim and esprocarb. Additionally, a metabolomics experiment on the central nervous system (CNS) of the freshwater snail Lymnaea stagnalis was conducted. The snails were exposed to the extract, the three most abundant chemicals individually, and a mixture of these. The extract disturbed more metabolic pathways than the three most abundant chemicals individually, indicating the contribution of other chemicals. Most pathways perturbed by the extract exposure overlapped with those related to exposure to neonicotinoids, like the polyamine metabolism involved in CNS injuries. Metabolomics for the straightforward comparison between a complex mixture and single compound toxicity is still challenging but, compared to traditional biotesting, is a promising tool due to its increased sensitivity.