Topics
Introduction to the R software
Participants will be introduced to R programming and to basic statistics in R, based on concrete toxicity test data. R is a language and environment for statistical computing and graphics (http://cran.r-project.org/). It provides a wide variety of statistical and graphical techniques. It is highly extensible through a great number of packages dedicated to specific topics. Participants will explore the most useful ones for ecotoxicity data analysis.DOSE-RESPONSE MODELING
There is today a large consensus recognizing LCx and ECx as an appropriate index of toxicity as far as data are sufficient and properly fitted with suitable models. Nevertheless, the choice of the appropriate model is not straightforward: both the deterministic and the stochastic parts must be suitably chosen in accordance with the type of experimental data (binary data e.g. for survival, quantitative continuous data e.g. for growth or count data e.g. for reproduction). Participants will be introduced to up-to-date methods, all included within the broader topic of generalized non-linear regression models.TOXICOKINETIC AND TOXICODYNAMIC MODELING
Toxicokinetic-toxicodynamic (TKTD) models allow to describe processes that lead to toxicity estimate over time at the organism level. More importantly TKTD models provide a conceptual framework to better understand the causes for variability in species sensitivity to the same compound as well as the causes for a different sensitivity of a same species to different compounds. The main benefit of TKTD models is their ability to bring knowledge about toxicity of compounds, sensitivity of organisms, organism recovery times and carry-over toxicity. Another advantage is their use to simulate the time course of toxicity. Participants will thus be introduced to what are TKTD models, distinguishing both TK and TD aspects, for what they can be employed and how to run one- or multi-compartments TK models, as well the TKTD model dedicated to survival over time, namely the GUTS framework.SPECIES SENSITIVITY DISTRIBUTIONS
Species-sensitivity distributions (SSD) analysis is a statistical approach that is employed to estimate either the concentration of a chemical that is hazardous to no more than p% of all species (the HCp) or, equivalently, the proportion p of species that are potentially affected by a given concentration of a chemical. SSD utilises sensitivity data as input, such as x% lethal or effective concentrations (ECx) for a collection of species, toxicity endpoints that can be left-, right- or interval-censored, e.g. when the uncertainty has been quantified. SSD analyses necessitate the fitting of a probability distribution, following the selection of an appropriate distribution law and an optimisation method. Participants will thus be introduced to these technical aspects based on practical case studies.BAYESIAN INFERENCE
Bayesian inference has showed its usefulness in ecotoxicology as a relevant alternative when estimating parameters of models, simultaneously fitted to different types of data. Within the Bayesian framework, parameter estimation requires three steps:1. model building, defining theoretical and stochastic links between variables depending on some parameters to be clearly identified;
2. prior distributions for model parameters reflecting the state of knowledge that is available about the mode before analysing the data;
3. posterior distributions of all parameters simultaneously computed based on the Bayes’ theorem principle combining priors and observations. Bayesian inference is very flexible and makes the use of generalized non-linear and TKTD models quite easy.
Participants will be trained step-by-step to both the theoretical and practical aspects of the Bayesian framework with the JAGS software (http://mcmc-jags.sourceforge.net/).