Accelerating the pace of ecotoxicological assessment using artificial intelligence

Song, Runsheng; Li, Dingsheng; Chang, Alexander; Tao, Mengya; Qin, Yuwei; Keller, Arturo A.; Suh, Sangwon

doi:10.1007/s13280-021-01598-8

Cited by 16 publications

(16 citation statements)

References 55 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The simplest train-test-split can be achieved by random sampling of data points, provided by us as totally random , which has been the main approach in previous work applying ML to ecotoxicology, and generally suffices for a well-balanced dataset without repeated experiments 31,56,57 . For our dataset with repeated experiments, i.e., data points coinciding in chemical, species, and experimental variables (Figure 2), the totally random approach has a high risk of data leakage and the associated overestimated model performances.…”

Section: Methodsmentioning

confidence: 99%

A Benchmark Dataset for Machine Learning in Ecotoxicology

Schür

Gasser

Pérez-Cruz

et al. 2023

Preprint

View full text Add to dashboard Cite

The use of machine learning for predicting ecotoxicological outcomes is promising, but underutilized. The curation of data with informative features requires both expertise in machine learning as well as a strong biological and ecotoxicological background, which we consider a barrier of entry for this kind of research. Additionally, model performances can only be compared across studies when the same dataset, cleaning, and splittings were used. Therefore, we provide ADORE, an extensive and well-described dataset on acute aquatic toxicity in three relevant taxonomic groups (fish, crustaceans, and algae). The core dataset describes ecotoxicological experiments and is expanded with phylogenetic and species-specific data on the species as well as chemical properties and molecular representations. Apart from challenging other researchers to try and achieve the best model performances across the whole dataset, we propose specific relevant challenges on subsets of the data and include datasets and splittings corresponding to each of these challenge as well as in-depth characterization and discussion of train-test splitting approaches.

show abstract

Section: Methodsmentioning

confidence: 99%

A Benchmark Dataset for Machine Learning in Ecotoxicology

Schür

Gasser

Pérez-Cruz

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The simplest train-test-split can be achieved by random sampling of data points, which has been the main approach in previous work applying ML to ecotoxicology and generally suffices for a well-balanced dataset without repeated experiments 12,30,31 . For the ADORE dataset with repeated experiments, i.e., data points coinciding in chemical, species, and experimental conditions, this approach has a high risk of data leakage and the associated overestimated model performances, as the same chemical as well as the same chemical-taxon pair are likely to appear in both the training and test set.…”

Section: Split Totally At Randommentioning

confidence: 99%

Machine learning-based prediction of fish acute mortality: Implementation, interpretation, and regulatory relevance

Gasser,

Schür,

Perez-Cruz

et al. 2024

Preprint

View full text Add to dashboard Cite

Regulation of chemicals requires knowledge of their toxicological effects on a large number of target species. Traditionally, this knowledge has been acquired throughin vivotesting. The recent effort to find alternatives based on machine learning, however, has not focused on guaranteeing transparency, comparability and reproducibility, which makes it difficult to assess advantages and disadvantages of these methods. Also, comparable baseline performances are needed. In this study, we trained regression models on the ADORE “t-F2F” challenge proposed in [Schüret al.,Nature Scientific data, 2023] to predict acute mortality, measured as LC50 (lethal concentration 50), of organic compounds on fishes. We trained LASSO, random forest (RF), XGBoost, Gaussian process (GP) regression models, and found a series of aspects that are stable across models: (i) using mass or molar concentrations does not affect performances; (ii) the performances are only weakly dependent on how a chemical is represented, but (iii) strongly on how the data is split. Overall, the tree-based models RF and XGBoost performed best and we were able to predict the log10-transformed LC50 with a root mean square error of 0.90, which corresponds to an order of magnitude on the original LC50 scale. On a local level, on the other hand, the models are not able to predict the toxicity of single chemicals accurately enough. Predictions for single chemicals are dominated by a few chemical properties and taxonomic traits are not captured sufficiently by the models. As a consequence, the models are not yet applicable in the regulatory process. Nevertheless, this work contributes to the ongoing discussions on how machine learning can be integrated in the regulatory process.Environmental significanceConventional environmental hazard assessment in its current form will not be able to adapt to the growing need for toxicity testing. Alternative methods, such as toxicity prediction through machine learning, could fulfill that need in an economically and ethically sound manner. Proper implementation, documentation, and the integration into the regulatory process are prerequisites for the usability and acceptance of these models.

show abstract

“…However, assessing the growing number of marketed chemicals across different consumer products, populations, and environments is increasingly challenging. , Characterizing chemical toxicity impacts, including aspects on environmental fate, exposure, and (eco-)toxicity effects, is essential across a wide range of chemical-related decision support tools, such as risk assessment , and screening, life cycle impact assessment (LCIA), , chemical footprinting, , chemical substitution, benchmarking chemical pollution against local-to-global boundaries, , and safe-and-sustainable-by-design (SSbD) assessments . The application of chemical-related decision support tools to the >100,000 marketed chemicals and the wide range of product uses is currently limited by a lack of structured, high-quality input data needed to characterize toxicity for millions of chemical–product combinations. , Obtaining new data from experimental tests is cost- and time-consuming, and confidential or nontransparent reporting hinder access to existing data. − To address data gaps, scientists have been developing quantitative structure–activity relationships (QSAR) for decades by creating quantitative links between chemical structures and various target properties, including input parameters for characterizing chemical toxicity. , With increasing data availability and computing power, QSAR evolved from simple regressions on small sets of congeneric compounds to applying advanced statistical and machine learning (ML) techniques on large chemical sets with diverse molecular structures, boosting their predictive performance and applicability for a broader realm of chemicals. , Several advanced chemical data prediction models are readily accessible through public modeling suites providing predictions for multiple chemical properties , and many more have been documented in the scientific literature for individual chemical properties, including dissociation constants, , root concentration factors, − and ecotoxicity end points. − While the development of ML-based approaches has been an active field of research, a systematic adaptation for chemical toxicity characterization is still limited. The main challenges relate to a lack of oversight into required input parameters which could support the systematic development of ML-based approaches and a lack of transparency about whether such approaches can robustly predict parameter...…”

Section: Introductionmentioning

confidence: 99%

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

von Borries,

Holmquist,

Kosnik

et al. 2023

Environ. Sci. Technol.

View full text Add to dashboard Cite

Machine Learning (ML) is increasingly applied to fill data gaps in assessments to quantify impacts associated with chemical emissions and chemicals in products. However, the systematic application of ML-based approaches to fill chemical data gaps is still limited, and their potential for addressing a wide range of chemicals is unknown. We prioritized chemical-related parameters for chemical toxicity characterization to inform ML model development based on two criteria: (1) each parameter's relevance to robustly characterize chemical toxicity described by the uncertainty in characterization results attributable to each parameter and (2) the potential for ML-based approaches to predict parameter values for a wide range of chemicals described by the availability of chemicals with measured parameter data. We prioritized 13 out of 38 parameters for developing ML-based approaches, while flagging another nine with critical data gaps. For all prioritized parameters, we performed a chemical space analysis to assess further the potential for ML-based approaches to predict data for diverse chemicals considering the structural diversity of available measured data, showing that ML-based approaches can potentially predict 8−46% of marketed chemicals based on 1−10% with available measured data. Our results can systematically inform future ML model development efforts to address data gaps in chemical toxicity characterization.

show abstract

Accelerating the pace of ecotoxicological assessment using artificial intelligence

Cited by 16 publications

References 55 publications

A Benchmark Dataset for Machine Learning in Ecotoxicology

A Benchmark Dataset for Machine Learning in Ecotoxicology

Machine learning-based prediction of fish acute mortality: Implementation, interpretation, and regulatory relevance

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

Contact Info

Product

Resources

About