Simultaneous prediction of toxicity of multiple chemicals to multiple species using multi‐dimensional functional relationships

Abstract

SUMMARYA theoretical model is developed for the estimation of the toxicity of a large number of chemicals to a group of species for which the log-toxicity values have strong linear relationships. The model is a multi-dimensional functional relationship (FR) that allows for many missing values. It depends on an unobserved variate (denoted by j in the text), with a value for each chemical; for each species, the expectations of the log-toxicity values are each linearly related to j. There are worthwhile gains in multi-dimensional over the two-dimensional FRs hitherto used. The model is applied to data on the toxicity of 51 chemicals to 10 species. It allows for species to have unequal variance about their expectations. There is good prediction of log-toxicity from measurements on other species where only 30% of the combinations were measured even when the taxa are not closely related. Among the uses for this technique are (i) to predict toxicity for combinations of species and chemicals that have not been measured, (ii) to suggest a suitable range of concentrations for a new bioassay, (iii) identifying specificity, that is, chemicals that have a different relative toxicity for a particular species.

Sensitivity of a Large and Representative Sample of Antarctic Marine Invertebrates to Metals

“…These relationships were, however, relatively weak, explaining 16% (Cu vs Zn) and 28% (Zn vs Cd) of the variability, and, as noted previously, there were a number of outliers (e.g., Antarctonemertes validum). Other studies have found more consistent correlations between chemicals in the sensitivity of species (r 2 = 0.66-0.90; Dyer et al 2008;Morton et al 2008 Hazardous concentrations (HCy to y% of the taxa tested) and log-normal distribution parameters (mean log, SD log) shown are best estimates (95% confidence intervals in parentheses). Log likelihood is an indication of the goodness of fit of the distribution.…”

A similarity‐index–based method to estimate chemical concentration limits protective for ecological communities

“…Expert opinion and Bayesian statistics refer to the use of expert opinion on the sensitivity of taxa calibrated against toxicity data using Bayesian statistics [10,11]. Statistical deviations involve estimation of the sensitivity of species based on their sensitivity to other substances relative to other species [9,13]. Abundance declines refer to estimation of the sensitivity of species based on the declines in their abundance in nature along a pollution gradient [2,12,[24][25][26].…”

“…Sensitivity of many nonstandard test species can also be estimated from expert opinion calibrated with Bayesian statistics against experimental conventional [10] or rapid [11] tests. The sensitivity of standard test species to a particular chemical can be derived from their relative sensitivity to other chemicals (statistical deviations) [9,13]. The sensitivity of sometimes abundant and widespread species can be estimated from abundance declines associated with increasing contamination and expressed in field-based SSDs [2,12,[24][25][26].…”

“…Recently, some modifications to SSDs [2,6,[9][10][11][12][13] have been suggested that should result in SSDs that better meet some of these assumptions. However, none of these modifications has attempted to address all assumptions of SSDs, and thus it will be controversial whether PCp values really protect p% of species.…”

Risk assessment using the species sensitivity distribution method: Data quality versus data quantity

“…Early protocols exclude all data that does not conform to high data quality standards . Recently, ecotoxicologists have argued for the inclusion of data conventionally regarded as being of lower quality, but including data from more species and a mix of species that better reflects the mix of species in specific environments . In addition, Bayesian statistics have recently been used to improve the community representativeness of sensitivity data included in SSDs .…”