Bio-QSARs 2.0: Unlocking a new level of predictive power for machine learning-based ecotoxicity predictions by exploiting chemical and biological information.

Practical, legal, and ethical reasons necessitate the development of methods to replace animal experiments. Computational techniques to acquire information that traditionally relied on animal testing are considered a crucial pillar among these so-called new approach methodologies. In this light, we recently introduced the Bio-QSAR concept for multispecies aquatic toxicity regression tasks. These machine learning models, trained on both chemical and biological information, are capable of both cross-chemical and cross-species predictions. Here, we significantly extend these models' applicability. This was realized by increasing the quantity of training data by a factor of approximately 20, accomplished by considering both additional chemicals and aquatic organisms. Additionally, variable test durations and associated random effects were accommodated by employing a machine learning algorithm that combines tree-boosting with mixed-effects modeling (i.e., Gaussian Process Boosting). We also explored various biological descriptors including Dynamic Energy Budget model parameters, taxonomic distances, as well as genus-specific traits and investigated the inclusion of mode-of-action information. Through these efforts, we developed Bio-QSARs for fish and aquatic invertebrates with exceptional predictive power (R squared of up to 0.92 on independent test sets). Moreover, we made considerable strides to make models applicable for a range of use cases in environmental risk assessment as well as research and development of chemicals. Models were made fully explainable by implementing an algorithmic multicollinearity correction combined with SHapley Additive exPlanations. Furthermore, we devised novel approaches for applicability domain construction that take feature importance into account. We are hence confident these models, which are available via open access, will make a significant contribution towards the implementation of new approach methodologies and ultimately have the potential to support "Green Chemistry" and "Green Toxicology".

Modeling temperature-dependent life-cycle toxicity of thiamethoxam in Chironomus riparius using a DEB-TKTD model.

The neonicotinoid insecticide thiamethoxam (TMX) is widely used to protect crops against insect pests. Despite some desirable properties such as its low toxicity to birds and mammals, concerns have been raised about its toxicity to non-target arthropods, including freshwater insects like chironomids. Whereas multiple studies have investigated chronic effects of neonicotinoids in chironomid larvae at standardized laboratory conditions, a better understanding of their chronic toxicity under variable temperatures and exposure is needed for coherent extrapolation from the laboratory to the field. Here, we developed a quantitative mechanistic effect model for Chironomus riparius, to simulate the species' life history under dynamic temperatures and exposure concentrations of TMX. Laboratory experiments at four different temperatures (12, 15, 20, 23 °C) and TMX concentrations between 4 and 51 µg/L were used to calibrate the model. Observed concentration-dependent effects of TMX in C. riparius included slower growth, later emergence, and higher mortality rates with increasing concentrations. Furthermore, besides a typical accelerating effect on the organisms' growth and development, higher temperatures further increased the effects associated with TMX. With some data-informed modeling decisions, most prominently the inclusion of a size dependence that makes larger animals more sensitive to TMX, the model was parametrized to convincingly reproduce the data. Experiments at both a constant (20 °C) and a dynamically increasing temperature (15-23 °C) with pulsed exposure were used to validate the model. Finally, the model was used to simulate realistic exposure conditions using two reference exposure scenarios measured in Missouri and Nebraska, utilizing a moving time window (MTW) and either a constant temperature (20 °C) or the measured temperature profiles belonging to each respective scenario. Minimum exposure multiplication factors leading to a 10% effect (EP10) in the survival at pupation, i.e., the most sensitive endpoint found in this study, were 25.67 and 21.87 for the Missouri scenario and 38.58 and 44.64 for the Nebraska scenario, when using the respective temperature assumptions. While the results illustrate that the use of real temperature scenarios does not systematically modify the EPx in the same direction (making it either more or less conservative when used as a risk indicator), the advantage of this approach is that it increases the realism and thus reduces the uncertainty associated with the model predictions.

Mechanistic population models for ecological risk assessment and decision support: The importance of good conceptual model diagrams.

The use of mechanistic population models as research and decision-support tools in ecology and ecological risk assessment (ERA) is increasing. This growth has been facilitated by advances in technology, allowing the simulation of more complex systems, as well as by standardized approaches for model development, documentation, and evaluation. Mechanistic population models are particularly useful for simulating complex systems, but the required model complexity can make them challenging to communicate. Conceptual diagrams that summarize key model elements, as well as elements that were considered but not included, can facilitate communication and understanding of models and increase their acceptance as decision-support tools. Currently, however, there are no consistent standards for creating or presenting conceptual model diagrams (CMDs), and both terminology and content vary widely. Here, we argue that greater consistency in CMD development and presentation is an important component of good modeling practice, and we provide recommendations, examples, and a free web app (pop-cmd.com) for achieving this for population models used for decision support in ERAs. Integr Environ Assess Manag 2024;00:1-9. © 2023 SETAC.

Density-dependent population regulation in freshwater fishes and small mammals: A literature review and insights for Ecological Risk Assessment.

The regulation of populations through density dependence (DD) has long been a central tenet of studies of ecological systems. As an important factor in regulating populations, DD is also crucial for understanding risks to populations from stressors, including its incorporation into population models applied for this purpose. However, study of density-dependent regulation is challenging because it can occur through various mechanisms, and their identification in the field, as well as the quantification of the consequences on individuals and populations, can be difficult. We conducted a targeted literature review specifically focusing on empirical laboratory or field studies addressing negative DD in freshwater fish and small rodent populations, two vertebrate groups considered in pesticide Ecological Risk Assessment (ERA). We found that the most commonly recognized causes of negative DD were food (63% of 19 reviewed fish studies, 40% of 25 mammal studies) or space limitations (32% of mammal studies). In addition, trophic interactions were reported as causes of population regulation, with predation shaping mostly small mammal populations (36% of the mammal studies) and cannibalism impacting freshwater fish (26%). In the case of freshwater fish, 63% of the studies were experimental (i.e., with a length of weeks or months). They generally focused on the individual-level causes and effects of DD, and had a short duration. Moreover, DD affected mostly juvenile growth and survival of fish (68%). On the other hand, studies on small mammals were mainly based on time series analyzing field population properties over longer timespans (68%). Density dependence primarily affected survival in subadult and adult mammal stages and, to a lesser extent, reproduction (60% vs. 36%). Furthermore, delayed DD was often observed (56%). We conclude by making suggestions on future research paths, providing recommendations for including DD in population models developed for ERA, and making the best use of the available data. Integr Environ Assess Manag 2023;00:1-12. © 2023 Syngenta Crop Protection. Integrated Environmental Assessment and Management published by Wiley Periodicals LLC on behalf of Society of Environmental Toxicology & Chemistry (SETAC).

Physiological variables in machine learning QSARs allow for both cross-chemical and cross-species predictions.

A major challenge in ecological risk assessment is estimating chemical-induced effects across taxa without species-specific testing. Where ecotoxicological data may be more challenging to gather, information on species physiology is more available for a broad range of taxa. Physiology is known to drive species sensitivity but understanding about the relative contribution of specific underlying processes is still elusive. Consequently, there remains a need to understand which physiological processes lead to differences in species sensitivity. The objective of our study was to utilize existing knowledge about organismal physiology to both understand and predict differences in species sensitivity. Machine learning models were trained to predict chemical- and species-specific endpoints as a function of both chemical fingerprints/descriptors and physiological properties represented by dynamic energy budget (DEB) parameters. We found that random forest models were able to predict chemical- and species-specific endpoints, and that DEB parameters were relatively important in the models, particularly for invertebrates. Our approach illuminates how physiological properties may drive species sensitivity, which will allow more realistic predictions of effects across species without the need for additional animal testing.

Population modeling to inform management and recovery efforts for lake sturgeon, Acipenser fulvescens.

Lake sturgeon (Acipenser fulvescens) populations have significantly declined across their historic range, in large part due to anthropogenic impacts that have likely been exacerbated by the life-history traits of this slow-growing and long-lived species. We developed a population model to explore how Contaminants of Emerging Concern (CECs) impact lake sturgeon populations. We explored how different physiological modes of action (pMoAs) of CECs impacted population abundance and recovery and how different simulated management actions could enable recovery. We first estimated the impacts on population abundance and recovery by comparing the trajectory of an unexposed population to a population that had been exposed to a CEC with a specific pMoA after the end of the exposure. We then predicted how different management actions would impact population recovery by comparing the trajectories of an unexposed population to an exposed population for which a management action started at a fixed time without discontinuation of the exposure. Our results predicted that the individual-level pMoA of CECs has an important impact on population-level effects because different stressor's pMoA impacts the life-history traits of sturgeon differently. For example, the feeding and reproduction pMoAs caused the strongest and weakest population declines, respectively. For the same reason, pMoA also impacted recovery. For example, recovery was delayed when the pMoA was growth, maintenance, or feeding, but it was immediate when the pMoA was reproduction. We found that management actions that increased the egg survival rate or the stocking of fingerlings resulted in faster and stronger recovery than management actions that increased the juvenile or adult survival rate. This result occurred because the first two management actions immediately impacted recruitment, whereas the impact was delayed for the last two. Finally, there was greater potential for recovery when management action targeted eggs and fingerlings because these life stages have lower natural survival rates. Integr Environ Assess Manag 2022;18:1597-1608. © 2022 Society of Environmental Toxicology & Chemistry (SETAC). This article has been contributed to by US Government employees and their work is in the public domain in the USA.

Differential retention contributes to racial/ethnic disparity in U.S. academia.

Several racial and ethnic identities are widely understood to be under-represented within academia, however, actual quantification of this under-representation is surprisingly limited. Challenges include data availability, demographic inertia and identifying comparison points. We use de-aggregated data from the U.S. National Science Foundation to construct a null model of ethnic and racial representation in one of the world's largest academic communities. Making comparisons between our model and actual representation in academia allows us to measure the effects of retention (while controlling for recruitment) at different academic stages. We find that, regardless of recruitment, failed retention contributes to mis-representation across academia and that the stages responsible for the largest disparities differ by race and ethnicity: for Black and Hispanic scholars this occurs at the transition from graduate student to postdoctoral researcher whereas for Native American/Alaskan Native and Native Hawaiian/Pacific Islander scholars this occurs at transitions to and within faculty stages. Even for Asian and Asian-Americans, often perceived as well represented, circumstances are complex and depend on choice of baseline. Our findings demonstrate that while recruitment continues to be important, retention is also a pervasive barrier to proportional representation. Therefore, strategies to reduce mis-representation in academia must address retention. Although our model does not directly suggest specific strategies, our framework could be used to project how representation in academia might change in the long-term under different scenarios.

A simulation-based evaluation of management actions to reduce the risk of contaminants of emerging concern (CECs) to walleye in the Great Lakes Basin.

Contaminants of emerging concern (CECs) are ubiquitous, present in complex chemical mixtures, and represent a threat to the Great Lake ecosystem. Mitigation strategies are needed to protect populations of key species, but knowledge about ecological and biological effects of CECs at the population level are limited. In this study, we combined laboratory data on CEC effects at the individual-level with in-situ CEC concentration data in a walleye (Sander vitreus) population model to simulate the effectiveness of different CEC mitigation strategies in the Maumee River and Lake Erie. We compared the effectiveness of moderate mitigation (50% reduction in exposure level) of an entire watershed versus intensive mitigation (reduction of exposure to a level that does not affect walleye) of single river sites for three CEC mixture scenarios (agricultural, urban, and combined). We also explored the impact of hypothetical chemical toxicokinetics (the time course of chemicals in walleye) on the relative effectiveness of the mitigation strategies. Our results suggest that when CECs impact fecundity, single-site mitigation is more effective when it focuses on spawning sites and nearby downstream sites that are substantially impaired. Our simulations also suggest that chemical toxicokinetics are important when evaluating single-site mitigation strategies, but that population characteristics, such as stage-specific mortality rate, are more important when evaluating watershed mitigation strategies. Results can be used to guide fisheries management, such as choosing habitat restoration sites, and identify key knowledge gaps that direct future research and monitoring.

Pop-guide: Population modeling guidance, use, interpretation, and development for ecological risk assessment.

The assimilation of population models into ecological risk assessment (ERA) has been hindered by their range of complexity, uncertainty, resource investment, and data availability. Likewise, ensuring that the models address risk assessment objectives has been challenging. Recent research efforts have begun to tackle these challenges by creating an integrated modeling framework and decision guide to aid the development of population models with respect to ERA objectives and data availability. In the framework, the trade-offs associated with the generality, realism, and precision of an assessment are used to guide the development of a population model commensurate with the protection goal. The decision guide provides risk assessors with a stepwise process to assist them in developing a conceptual model that is appropriate for the assessment objective and available data. We have merged the decision guide and modeling framework into a comprehensive approach, Population modeling Guidance, Use, Interpretation, and Development for Ecological risk assessment (Pop-GUIDE), for the development of population models for ERA that is applicable across regulatory statutes and assessment objectives. In Phase 1 of Pop-GUIDE, assessors are guided through the trade-offs of ERA generality, realism, and precision, which are translated into model objectives. In Phase 2, available data are assimilated and characterized as general, realistic, and/or precise. Phase 3 provides a series of dichotomous questions to guide development of a conceptual model that matches the complexity and uncertainty appropriate for the assessment that is in concordance with the available data. This phase guides model developers and users to ensure consistency and transparency of the modeling process. We introduce Pop-GUIDE as the most comprehensive guidance for population model development provided to date and demonstrate its use through case studies using fish as an example taxon and the US Federal Insecticide Fungicide and Rodenticide Act and Endangered Species Act as example regulatory statutes. Integr Environ Assess Manag 2021;17:767-784. © 2020 SETAC. This article has been contributed to by US Government employees and their work is in the public domain in the USA.

A Review of Key Features and Their Implementation in Unstructured, Structured, and Agent-Based Population Models for Ecological Risk Assessment.

Population models can provide valuable tools for ecological risk assessment (ERA). A growing amount of work on model development and documentation is now available to guide modelers and risk assessors to address different ERA questions. However, there remain misconceptions about population models for ERA, and communication between regulators and modelers can still be hindered by a lack of clarity in the underlying formalism, implementation, and complexity of different model types. In particular, there is confusion about differences among types of models and the implications of including or ignoring interactions of organisms with each other and their environment. In this review, we provide an overview of the key features represented in population models of relevance for ERA, which include density dependence, spatial heterogeneity, external drivers, stochasticity, life-history traits, behavior, energetics, and how exposure and effects are integrated in the models. We differentiate 3 broadly defined population model types (unstructured, structured, and agent-based) and explain how they can represent these key features. Depending on the ERA context, some model features will be more important than others, and this can inform model type choice, how features are implemented, and possibly the collection of additional data. We show that nearly all features can be included irrespective of formalization, but some features are more or less easily incorporated in certain model types. We also analyze how the key features have been used in published population models implemented as unstructured, structured, and agent-based models. The overall aim of this review is to increase confidence and understanding by model users and evaluators when considering the potential and adequacy of population models for use in ERA. Integr Environ Assess Manag 2021;17:521-540. © 2020 SETAC.

Guidance for Developing Amphibian Population Models for Ecological Risk Assessment.

Despite widespread acceptance of the utility of population modeling and advocacy of this approach for a more ecologically relevant perspective, it is not routinely incorporated in ecological risk assessments (ERA). A systematic framework for situation-specific model development is one of the major challenges to broadly adopting population models in ERA. As risk assessors confront the multitude of species and chemicals requiring evaluation, an adaptable stepwise guide for model parameterization would facilitate this process. Additional guidance on interpretation of model output and evaluating uncertainty would further contribute to establishing consensus on good modeling practices. We build on previous work that created a framework and decision guide for developing population models for ERA by focusing on data types, model structure, and extrinsic stressors relevant to anuran amphibians. Anurans have a unique life cycle with varying habitat requirements and high phenotypic plasticity. These species belong to the amphibian class, which is facing global population decline in large part due to anthropogenic stressors, including chemicals. We synthesize information from databases and literature relevant to amphibian risks to identify traits that influence exposure likelihood, inherent sensitivity, population vulnerability, and environmental constraints. We link these concerns with relevant population modeling methods and structure in order to evaluate pesticide effects with appropriate scale and parameterization. A standardized population modeling approach, with additional guidance for anuran ERA, offers an example method for quantifying population risks and evaluating long-term impacts of chemical stressors to populations. Integr Environ Assess Manag 2020;16:223-233. © 2019 SETAC.

Similar individual-level responses to stressors have different population-level consequences among closely related species of trout.

In this paper, we applied an individual-based model to study the population-level impacts of sub-lethal stressors affecting the metabolic pathways of three closely related trout species: Oncorhynchus mykiss (rainbow trout, RT), Salmo trutta (brown trout, BT) and Oncorhynchus calrki stomias (greenback cutthroat trout, GCT). Both RT and BT are well-studied species, and the former is widely used as a standard cold-water test species. These species are known to outcompete GCT, which is listed as threatened under the US Endangered Species Act. Our goal was to understand the extent to which stressor effects, which are often measured at the individual level, on taxonomically-related (i.e., surrogate) species can be informative of impacts on population dynamics in species that cannot be tested (e.g., listed species). When comparing stressor effects among species, we found that individual-level responses to each stressor were qualitatively comparable. Individual lengths and number of eggs decreased by similar percentages with respect to baseline, even if small quantitative differences were present depending on the physiological mode of action of the stressor. Individual-level effects in GCT were slightly greater when ingestion efficiency decreased, whereas effects in GCT and RT were greater when maintenance costs increased, and effects in BT were slightly greater when costs of growth increased. In contrast, results at the population level differed markedly among species with GCT the most impacted by sub-lethal stress effects on individual metabolism. Our findings suggest that using non-listed species to assess the risks of stressors to listed species populations may be misleading, even if the species are closely related and show similar individual-level responses. Mechanistic population models that incorporate species life history and ecology can improve inter-species extrapolation of stressor effects.

A mechanistic individual-based model of the feeding processes for Oikopleura dioica.

A mechanistic physiological model of the appendicularian Oikopleura dioica has been built to represent its three feeding processes (filtration, ingestion and assimilation). The mathematical formulation of these processes is based on laboratory observations from the literature, and tests different hypotheses. This model accounts for house formation dynamics, the food storage capacity of the house and the gut throughput dynamics. The half-saturation coefficient for ingestion resulting from model simulations is approximately 28 [Formula: see text] and is independent of the weight of the organism. The maximum food intake for ingestion is also a property of the model and depends on the weight of the organism. Both are in accordance with data from the literature. The model also provides a realistic representation of carbon accumulation within the house. The modelled half-saturation coefficient for assimilation is approximately 15 [Formula: see text] and is also independent of the weight of the organism. Modelled gut throughput dynamics are based on faecal pellet formation by gut compaction. Model outputs showed that below a food concentration of 30 [Formula: see text], the faecal pellet weight should represent a lower proportion of the body weight of the organism, meaning that the faecal pellet formation is not driven by gut filling. Simulations using fluctuating environmental food availability show that food depletion is not immediately experienced by the organism but that it occurs after a lag time because of house and gut buffering abilities. This lag time duration lasts at least 30 minutes and can reach more than 2 hours, depending on when the food depletion occurs during the house lifespan.