Examining predictors of chemical toxicity in freshwater fish using the random forest technique

Abstract

Chemical pollution is one of the main issues globally threatening the enormous biodiversity of freshwater ecosystems. The toxicity of substances depends on many factors such as the chemical itself, the species affected, environmental conditions, exposure duration, and concentration. We used the random forest technique to examine the factors that mediate toxicity in a set of widespread fishes and analyses of covariance to further assess the importance of differential sensitivity among fish species. Among 13 variables, the 5 most important predictors of toxicity with random forests were, by order of importance, the chemical substance itself (i.e., Chemical Abstracts Service number considered as a categorical factor), octanol-water partition coefficient (log P), pollutant prioritization, ecological structure-activity relationship (ECOSAR) classification, and fish species for 50% lethal concentrations (LC) and the chemical substance, fish species, log P, ECOSAR classification, and water temperature for no observed effect concentrations (NOECs). Fish species was a very important predictor for both endpoints and with the two contrasting statistical techniques used. Different fish species displayed very different relationships with log P, often with different slopes and with as much importance as the partition coefficient. Therefore, caution should be exercised when extrapolating toxicological results or relationships among species. In addition, further research is needed to determine species-specific sensitivities and unravel the mechanisms behind them.

Predicting Chemical Hazard across Taxa through Machine Learning

“…These observations are consistent with Ref. [12], where the species is identified as the most important taxonomic descriptor, and the LogP as the most important chemical. However, collinearity must be taken into account when interpreting Fig.…”

“…For example, the importance of the fish family would likely increase if the species was not used. This means that we expect that lower taxonomic ranks have a higher importance, but that the lower importance of higher taxonomic ranks does not make them irrelevant, as it was previously alluded [12]. A similar reasoning also applies chemical descriptors.…”

“…However, it is difficult to state whether these differences are statistically significant, and, if so, whether it should be attributed to the smaller size of the RT and BS datasets. Tuulaikhuu et al studied the Eco-Tox database, which contained enough data to allow for using taxonomy as a training feature [12]. They trained random forests to predict acute fish toxicity data (LC50), obtaining an R 2 = 0.85.…”

Key factors explaining critical swimming speed in freshwater fish: a review and statistical analysis for Iberian species

“…We used random forest (RF) 118 , as implemented in the package “party” 119 of the R software 120 , to analyze which of the six predictors best explained U crit . RF is a machine-learning technique that is frequently used because of their advantages, including computational efficiency on large databases with many correlated predictors, the provision of estimates of variable importance, the ability to impute missing data while maintaining accuracy, and the handling of non-linearities and interactions 118 , 121 . Specifically, RF computed with package “party” has the advantage of providing unbiased variable selection compared to other software packages, because it is more accurate when predictors are correlated and vary in their measurement scale or number of categories 122 , 123 .…”

The Multivariate Analysis of Foliar Fertilization by Using Random Forest for Improving Alliin Concentration of Garlic (<i>Allium sativum l.</i>)

“…RF is an ensemble machine learning algorithm for classification and regression which is constructed by a multitude of full depth decision trees without pruning [17], it is a nonparametric tree-based approach that merges the ideas of adaptive nearest neighbors with bagging for effective data adaptive inference [18]. RF has been successfully employed in dealing with various biological prediction problems [19][20][21][22], such microarray data microarray data classification [23][24][25], SNPs (single nucleotide polymorphisms) [26], chemical toxicity prediction [27], DNA-binding proteins identification [28,29], it is regarded as a useful alternative to capture the complex interaction effects among the many factors.…”