Goals

The DEMETRA models have been developed starting from very high quality data, using advanced computerised modelling techniques and validated with a range of methods.
Five models have been developed, each of them composed of a number of sub-models, to improve the predictive power and reliability.
Only data produced according to official guidelines have been used. The DEMETRA models are based on toxicity data from the US EPA, German BBA and the European project SEEM. Toxicity values from these three databases have been compared and verified, to increase reliability.The chemical names, formulas and structures have also been individually checked by more than one partner under a strict quality-control to ensure accuracy.
Thousands of chemical descriptors have been calculated using several specialist software programs. However, only the most useful chemical descriptors were selected for use in the final models.For modelling purposes, several techniques have been tested, both to classify chemicals and to predict toxicity as continuous values. Linear and non-linear mathematical methods have been used, such as PLS, neural networks, fuzzy logic, machine learning.
Hundreds of models have been developed; however, only the best ones were kept and combined within a hybrid model for each endpoint.