DEMETRA Models

Publicly available models may be open source or not. Another difference is that some models require calculation of the chemical descriptors as a separate task, while others simply need the chemical structure. Examples are the EPA [11] and DEMETRA models [20],... 

The DEMETRA models have been optimized to minimize the number of false negatives in order to achieve a safer regulatory application. Conversely, evaluation using only squared statistical parameters (such as R2), which deal with false positives and negatives in the same manner, is typical in QSAR modeling. 

C. Porcelli et al., Regulatory perspectives in the use and validation of QSAR. A case study DEMETRA model for daphnia toxicity. Environ. Sci. Technol. 42, 491 196 (2008)... 

DEMETRA models are based on hybrid techniques. Each model for trout, daphnia, quail, or bee is composed of a number of submodels. A hybrid model integrates, in an intelligent way, the results of the submodels to achieve better prediction values capable of reducing false negatives. The regulator can see the minimal and maximal values of the individual models, should the regulator decide to use the lowest predicted value. Another innovative feature of DEMETRA is that its models are freely available (http //www.demetra-tox. net) for wide use, and the user has only to calculate 2D descriptors. 

Another aspect is that multiple QSAR models can exist, and indeed this is likely. The consequences of such a complex picture, with many endpoints and models based on different approaches, are that it is not realistic to have a single approach, but the models have to be modulated considering these issues. Different approaches can be used for the same target and they improve the reliability of the prediction. In some cases, they have been combined into a unified system to improve overall results. This is the case of the models for regulatory purposes developed within the project DEMETRA (see below). 

Another difference in the models is that some are fixed, focused for a certain target, and optimized for that. It is the case of many of the commercial programs and of the public models from EPA and DEMETRA. 

The DEMETRA project [20, 26] represents the first important case of a European project to develop QSAR models for regulatory purposes. DEMETRA developed five free models to determine the ecotoxicity of pesticides using endpoints that include trout, daphnia, bees, and quails (oral and dietary exposure). Since the target was to develop models for the user, these five endpoints were decided by them, not by the developers. 

We notice that in practically all cases, a QSAR model was developed having only considered the data availability, without the other assessment evaluation carried out for the DEMETRA project. 

DEMETRA clearly stated the requirements for the data to be used because the models were based on the Pesticide Directive, and only the data produced according to the protocols defined in the directive were used. 

Regulators and stakeholders within DEMETRA project defined that an accepted uncertainty of the toxicity experimental data was a factor of 4. If the data uncertainty for the same chemical was higher, the chemical was not used for the model in the training set. Here, we notice that QSAR studies do not involve looking at the uncertainty of the property/toxicity value, thereby highlighting the difference between typical and regulatory models. 

The full details of the DEMETRA project are available, with full description and explanation [26] and the models and structures are also available [20]. 

The approach of CAESAR is quite similar to that of DEMETRA. So far good results have been obtained for the bioconcentration factor (BCF) in fish, superior to those of other models. Models have been tested with an external validation set. The model gives as prediction the BCF as continuous value, but it has been optimized to reduce false negatives. In the specific case of the REACH legislation, which is the target of the project, bioaccumulative chemicals are defined if the BCF value is above 3.3 in logarithmic unit. This shows another example of the specificity of the models, because different threshold may apply in other countries. 

The EC funded project CHEMOMENTUM will also implement QSAR models for pesticides and industrial chemicals, taken from the DEMETRA and CAESAR projects, and presented in a more user-friendly form [29]. 

E. Benfenati, The specificity of the QSAR models for regulatory purposes the example of the DEMETRA project. SAR QSAR Environ. Res. 18, 209-220 (2007)... 

Very recently the EC funded project DEMETRA (http //www.demetra-tox. net) developed a series of QSAR models for the prediction of toxicity of pesticides toward five endpoints trout, daphnia, quail (oral and dietary exposure), and bee [80]. This project introduced a number of innovative issues, compared to previous QSAR models. The target of the project was to develop models for pesticides to be used for regulatory purposes in accord with European legislation. A questionnaire was distributed to a great many end-user to identify their needs. The endpoints to be modeled were chosen from among those defined in writing by the end-user, and not by the modeler, in order to make the models as useful as possible. This attention to the needs of the end-users is unique in the use of QSAR for ecotoxicity prediction. Other novel... 

DEMETRA includes specific confidence restrictions capable of identifying outliers. The regulator can chose the confidence level for a factor of 50 or 10 and then can apply, or not, some restriction rules [80]. This way models with... 

---