The five Demetra models are the final results of a multi-step process started with the selection of the five end-points and ended with the release of the applet. The main steps of this process are:

• The selection of the Five Endpoints and evaluation of Toxicity Data;
• The description of the Chemical Information (Chemical Structures and Chemical Descriptors);
• The building of the Hybrid Models;
• The Demetra Tool;
• The analysis of the Performances of the models and outliers.

In order to build QSAR for regulatory purpose the Demetra models have been carefully checked and prepared according to quality criteria based on the specific intended application for regulatory purposes using a novel approach. In particular we considered the European pesticide directive, the officials guidelines for experimental data and the OECD principles for QSAR.

Figure 1 - Trout.

Predicted vs. experimental values for trout high confidence hybrid model. The dotted lines shows a confidentiality of a factor of 10 in case of false negatives and a factor of 50 in case of false positives. With permission from Quantitative Structure-Activity Relationships (QSAR) for Pesticide Regulatory Purposes. Ed. Emilio Benfenati. Elsevier, Amsterdam. In press.

Figure 2 – Daphnia.

Figure 3 – Quail, oral exposure.

Figure 4 – Quail, dietary exposure.

Predicted vs. experimental values of the hybrid model for quail, dietary exposure. Dotted lines indicate a factor of 50 in mg/kg. With permission from Quantitative Structure-Activity Relationships (QSAR) for Pesticide Regulatory Purposes. Ed. Emilio Benfenati. Elsevier, Amsterdam. In press.

Figure 5. – Bee.

Predicted vs. experimental values for bee high confidence hybrid model. The dotted lines shows a confidentiality of a factor of 10 in case of false negatives and a factor of 50 in case of false positives. With permission from Quantitative Structure-Activity Relationships (QSAR) for Pesticide Regulatory Purposes. Ed. Emilio Benfenati. Elsevier, Amsterdam. In press.