The EFSA ‘Guidance on tiered risk assessment for edge-of-field surface waters’ underscores the importance of in silico models to support the pesticide risk assessment. The aim of this work was to use in silico models starting from an available, structured and harmonized pesticide dataset that was developed for different purposes, in order to stimulate the use of QSAR models for risk assessment. The present work focuses on the development of a set of in silico models, developed to predict the aquatic toxicity of heterogeneous pesticides with incomplete/unknown toxic behavior in the water compartment. The generated models have good fitting performances (R2: 0.75–0.99), they are internally robust (Q2loo: 0.66–0.98) and can handle up to 30% of perturbation of the training set (Q2 lmo: 0.64–0.98). The absence of chance correlation was guaranteed by low values of R2 calculated on scrambled responses (R2 Yscr: 0.11–0.38). Different statistical parameters were used to quantify the external predictivity of the models (CCCext: 0.73–0.91, Q2 ext-Fn: 0.53–0.96). The results indicate that all the best models are predictive when applied to chemicals not involved in the models development. In addition, all models have similar accuracy both in fitting and in prediction and this represents a good degree of generalization. These models may be useful to support the risk assessment procedure when experimental data for key species are missing or to create prioritization lists for the general a priori assessment of the potential toxicity of existing and new pesticides which fall in the applicability domain.
Application of chemometric methods and QSAR models to support pesticide risk assessment starting from ecotoxicological datasets
Moretto A.;
2020
Abstract
The EFSA ‘Guidance on tiered risk assessment for edge-of-field surface waters’ underscores the importance of in silico models to support the pesticide risk assessment. The aim of this work was to use in silico models starting from an available, structured and harmonized pesticide dataset that was developed for different purposes, in order to stimulate the use of QSAR models for risk assessment. The present work focuses on the development of a set of in silico models, developed to predict the aquatic toxicity of heterogeneous pesticides with incomplete/unknown toxic behavior in the water compartment. The generated models have good fitting performances (R2: 0.75–0.99), they are internally robust (Q2loo: 0.66–0.98) and can handle up to 30% of perturbation of the training set (Q2 lmo: 0.64–0.98). The absence of chance correlation was guaranteed by low values of R2 calculated on scrambled responses (R2 Yscr: 0.11–0.38). Different statistical parameters were used to quantify the external predictivity of the models (CCCext: 0.73–0.91, Q2 ext-Fn: 0.53–0.96). The results indicate that all the best models are predictive when applied to chemicals not involved in the models development. In addition, all models have similar accuracy both in fitting and in prediction and this represents a good degree of generalization. These models may be useful to support the risk assessment procedure when experimental data for key species are missing or to create prioritization lists for the general a priori assessment of the potential toxicity of existing and new pesticides which fall in the applicability domain.Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.