QSAR‐Based Estimation of Species Sensitivity Distribution Parameters: An Exploratory Investigation

Abstract

Ecological risk assessments are hampered by limited availability of ecotoxicity data. The present study aimed to explore the possibility of deriving species sensitivity distribution (SSD) parameters for nontested compounds, based on simple physicochemical characteristics, known SSDs for data‐rich compounds, and a quantitative structure–activity relationship (QSAR)‐type approach. The median toxicity of a data‐poor chemical for species assemblages significantly varies with values of the physicochemical descriptors, especially when based on high‐quality SSD data (from either acute median effect concentrations or chronic no‐observed‐effect concentrations). Beyond exploratory uses, we discuss how the precision of QSAR‐based SSDs can be improved to construct models that accurately predict the SSD parameters of data‐poor chemicals. The current models show that the concept of QSAR‐based SSDs supports screening‐level evaluations of the potential ecotoxicity of compounds for which data are lacking. Environ Toxicol Chem 2019;38:2764–2770. © 2019 The Authors. Environmental Toxicology and Chemistry published by Wiley Periodicals, Inc. on behalf of SETAC

Concerns about Reproducibility, Use of the Akaike Information Criterion, and Related Issues in Hoondert et al. 2019

“…Given the difficulty of and limitations in generating ecotoxicological data, seeking quantitative structure-activity relationship (QSAR) approaches to estimating SSDs is valuable. The recent article by Hoondert et al (2019) is pioneering in that they developed QSAR-based models to estimate means and standard deviations (SDs) of SSDs based on acute and chronic toxicity data. Hoondert et al (2019) selected a total of 4 models to predict means and SDs of acute and chronic SSDs and then discussed the importance of the predictors included in the models.…”

“…The recent article by Hoondert et al (2019) is pioneering in that they developed QSAR-based models to estimate means and standard deviations (SDs) of SSDs based on acute and chronic toxicity data. Hoondert et al (2019) selected a total of 4 models to predict means and SDs of acute and chronic SSDs and then discussed the importance of the predictors included in the models. However, we failed to reproduce their model selection results (i.e., Table 3 of Hoondert et al [2019]) because their description of the method is not sufficiently detailed.…”

“…Hoondert et al (2019) selected a total of 4 models to predict means and SDs of acute and chronic SSDs and then discussed the importance of the predictors included in the models. However, we failed to reproduce their model selection results (i.e., Table 3 of Hoondert et al [2019]) because their description of the method is not sufficiently detailed. Because they state that "the most parsimonious models were…selected…based on the corrected Akaike information criterion as well as the adjusted R 2 ," we infer that they used both the corrected Akaike information criterion (AICc; Burnham et al [2011]) and adjusted R 2 values for model selection.…”

“…Returning to our major concern, when we analyzed the original data available in the Supplemental Data, we found that the best models with the smallest AICc values were inconsistent with the models selected in Hoondert et al (2019; e.g., see Table 1 for the SSD mean for acute data). Additionally, the models selected by Hoondert et al (2019) for estimating mean and SD of SSDs based on chronic toxicity data had delta values (differences in AICc values between the best models and corresponding models) of >20 (Supplemental Data; see below for more details on the interpretation of AICc differences). These delta values suggest that the models selected had essentially no empirical support (Burnham et al 2011).…”

“…If we examine the list of all models with a delta value <10 (Table 1), we see that 2 parameters, solubility and functional groups, were included in all the models. Their inclusion suggests that they are important for predicting the mean of the acute SSD, as noted by Hoondert et al (2019). Molecular weight and vapor pressure were often included in the top 5 models with delta values ≤2.1, but biodegradability was not (Table 1).…”