ESR1 Inés Rodríguez Leal
Project: In silico approaches for the identification of natural toxins relevant to water quality
Principal supervisor: Prof. Matthew Macleod
Intro to project
Natural toxins constitute a potential risk to water supplies in Europe. Only a few systematic risk assessments of individual natural toxins have been conducted in Europe. There is a need to update these risk assessments to reflect possible effects of climate change, especially changes in the distribution of agricultural plants throughout the continent and increasing prevalence of monoculture farming. Furthermore, screening-level assessment of many natural toxins that have been identified but not fully assessed is needed. This project aims to develop methods to prioritize natural toxins in water, mainly through the study of physico-chemical properties using QSAR methods, and environmental modeling. We envisage that the outcomes of our approach will be useful to guide more detailed toxicity testing and property measurements, and the development of monitoring programs and regulations to ensure safe drinking water in Europe.
QSAR predictions are based on molecular structure and hence should be considered carefully when applied to chemicals that are structurally different from those that were used to develop the model. The establishment of an applicability domain of the models provides a range of chemicals where the predictions are expected to be reliable, based on their molecular similarity. Therefore, the first step has been to establish the applicability domain of selected EPISuite prediction models towards a preliminary list of natural toxins to determine the reliable predicted properties. Such predictions will be used to build a database with prioritized toxins according to their persistence and mobility. Methods and results were presented in a poster session at the SETAC Rome conference in May 2018.
A limitation of existing EPISuite prediction models is that natural toxins are not present in the model training sets. These models could produce large and unexpected errors when applied to natural toxins that have structural features that are not present in the training sets, which would be the QSAR modeling equivalent of a black swan event. To begin to address this possibility, we developed a new QSAR model using experimental Koc values for natural toxins from the NaToxAq consortium and for other chemicals collected from the literature. Predictors to build the model were extracted and analyzed with R, applying stepwise linear regression and partial least squares methods to achieve models that allow to better predict Koc values for natural toxins. Results were presented in the NaToxAq online conference in September 2020.