Final Report Summary - CAESAR (Computer-Assisted Evaluation of industrial chemical Substances According to Regulations)
'Registration, evaluation, authorisation and restriction of chemical substances' (REACH) is the new European Community regulation on chemicals and their safe use, aiming to protect human health and the environment. In order to reduce the increasing of animal tests, which were necessary in order to provide all the required data for new products' commercialisation, the REACH legislation established the use of alternative methods, like in vitro or in silico methods.
The CAESAR project aimed to develop in silico models to predict the biological activity of the chemicals for five endpoints selected between those that required extensive animal experiments, in order to reduced in-vivo experiments and thus required costs. CAESAR produced freely available quantitative structure-activity relationship (QSAR) models for bioconcentration factor (BCF), skin sensitisation, mutagenicity, carcinogenicity and developmental toxicity, or teratogenicity. QSAR approach allowed for the prediction of the missing property on the basis of the chemical structure, while it permitted simultaneous testing of many compounds. The availability of the produced models served the objective of achieving wide applicability and acceptability of the models by stakeholders for decision support.
In silico models were not widely acceptable prior to CAESAR initiation, due to the lack of reliable databases of experimental data and the limited applicability domain of such models. In order to overcome the first limitation, the availability of extensive, high quality experimental data to serve as training set for the predictive models' creation was among the criteria for the selection of the endpoints examined by the project.
In case of skin sensitisation two complementary approaches were investigated; a global one to discriminate sensitisers and non-sensitisers and a local one which identified a mechanistic based category formation coupled with a read-across approach within each category. Two complementary methodologies were also necessary to address the carcinogenicity problem, namely a regression and a continuous approach.
Since QSAR models predicted the activity of a compound based on its chemical structure, the use of correct structures was fundamental. As a result the chemical structures of all the compounds in the utilised data sets were pre-processed in order to obtain data sets of high quality. The available compounds were subsequently divided in a training set, including the data majority, and a test set, which was used to evaluate the developed models' performance. Two-dimensional chemical descriptors were selected to represent compounds' chemical structures because of their simplicity and sufficient performance, which ensured reproducibility of the results.
Extensive calculations and analyses were performed during the models' development; however, they resulted in the production of simplified, easy-to-use tools, which incorporated the optimal, yet practical, solutions. The produced models became available to users through the project website, which also contained ready-to-use algorithms for chemical descriptors' calculations.
The developed models performed better or comparably to pre-existing ones as it occurred by experimental tests. However, in the carcinogenicity case, the QSAR model could not totally replace animal testing. Nevertheless, it could contribute to the evaluation of carcinogenicity as an additional source of information. Moreover, the relatively limited number of compounds which were used for the creation of the model of developmental toxicity resulted in the recommendation that the end product was used only as support information.
An additional tool, determining the chemicals of a training set which were similar to the compound of interest and providing a similarity index value was developed, so that predictions for similar compounds could be easily evaluated. Apart from that, other activities which were not initially planned, such as the development of numerous models for a single endpoint and the development of adequate tools so that the user did not have to calculate chemical compounds proved that CAESAR overcame its initial objectives.
Collaborations between experts of the field were promoted in order to improve knowledge exchange and achieve qualitative results. Knowledge dissemination activities included, apart from the project website, publications in scientific journals and participations in conferences, along with the organisation of specialised workshops focusing on the project activities.
CAESAR approach was to develop the same model for both industry and regulators, so that it was transparent and reproducible and both sectors could benefit from its use. All produced models were in accordance with the quality protocol and could be used for regulatory purposes. CAESAR products fulfilled REACH requirements since they included all the necessary factors to increase the acceptance of QSAR models. In addition, the models covered REACH most important endpoints, for which suitable models did not exist. Finally, guidelines for selecting the quality of the properties that were used as input in the models were also generated.
The CAESAR project aimed to develop in silico models to predict the biological activity of the chemicals for five endpoints selected between those that required extensive animal experiments, in order to reduced in-vivo experiments and thus required costs. CAESAR produced freely available quantitative structure-activity relationship (QSAR) models for bioconcentration factor (BCF), skin sensitisation, mutagenicity, carcinogenicity and developmental toxicity, or teratogenicity. QSAR approach allowed for the prediction of the missing property on the basis of the chemical structure, while it permitted simultaneous testing of many compounds. The availability of the produced models served the objective of achieving wide applicability and acceptability of the models by stakeholders for decision support.
In silico models were not widely acceptable prior to CAESAR initiation, due to the lack of reliable databases of experimental data and the limited applicability domain of such models. In order to overcome the first limitation, the availability of extensive, high quality experimental data to serve as training set for the predictive models' creation was among the criteria for the selection of the endpoints examined by the project.
In case of skin sensitisation two complementary approaches were investigated; a global one to discriminate sensitisers and non-sensitisers and a local one which identified a mechanistic based category formation coupled with a read-across approach within each category. Two complementary methodologies were also necessary to address the carcinogenicity problem, namely a regression and a continuous approach.
Since QSAR models predicted the activity of a compound based on its chemical structure, the use of correct structures was fundamental. As a result the chemical structures of all the compounds in the utilised data sets were pre-processed in order to obtain data sets of high quality. The available compounds were subsequently divided in a training set, including the data majority, and a test set, which was used to evaluate the developed models' performance. Two-dimensional chemical descriptors were selected to represent compounds' chemical structures because of their simplicity and sufficient performance, which ensured reproducibility of the results.
Extensive calculations and analyses were performed during the models' development; however, they resulted in the production of simplified, easy-to-use tools, which incorporated the optimal, yet practical, solutions. The produced models became available to users through the project website, which also contained ready-to-use algorithms for chemical descriptors' calculations.
The developed models performed better or comparably to pre-existing ones as it occurred by experimental tests. However, in the carcinogenicity case, the QSAR model could not totally replace animal testing. Nevertheless, it could contribute to the evaluation of carcinogenicity as an additional source of information. Moreover, the relatively limited number of compounds which were used for the creation of the model of developmental toxicity resulted in the recommendation that the end product was used only as support information.
An additional tool, determining the chemicals of a training set which were similar to the compound of interest and providing a similarity index value was developed, so that predictions for similar compounds could be easily evaluated. Apart from that, other activities which were not initially planned, such as the development of numerous models for a single endpoint and the development of adequate tools so that the user did not have to calculate chemical compounds proved that CAESAR overcame its initial objectives.
Collaborations between experts of the field were promoted in order to improve knowledge exchange and achieve qualitative results. Knowledge dissemination activities included, apart from the project website, publications in scientific journals and participations in conferences, along with the organisation of specialised workshops focusing on the project activities.
CAESAR approach was to develop the same model for both industry and regulators, so that it was transparent and reproducible and both sectors could benefit from its use. All produced models were in accordance with the quality protocol and could be used for regulatory purposes. CAESAR products fulfilled REACH requirements since they included all the necessary factors to increase the acceptance of QSAR models. In addition, the models covered REACH most important endpoints, for which suitable models did not exist. Finally, guidelines for selecting the quality of the properties that were used as input in the models were also generated.
