In the second 1.5 years of the project, the emphasis of the activities was on further operationalization of tools, data, models, and non-testing approaches for future hazard, safety and risk assessment for chemicals within the four “CADASTER-classes”, according to the REACH philosophy. In practice this implied finalization of the project infrastructure, additional collection of data and models, model development, investigation of the means of implementation of models in legislation, investigation of the possibilities of application and integration of modelling tools in probabilistic risk assessment. Progress of the project is as planned, and the deliverables and milestones agreed upon have been met, in some cases additional activities have been initiated, and additional scientific reports have been published (contributions to workshops and symposia, publications in peer-reviewed journals). All products are available from the CADASTER website.

The main activities in the first three years of CADASTER focused on:

1. Generating experimental data relevant for hazard assessment for chemicals from within the chemical classes identified within CADASTER: the SIDS end points;
2. Developing and validating models for predicting fate and effect properties of the chemicals from within the chemical classes identified within CADASTER;
3. Performing a gap analysis regarding the most essential data and models that are lacking for performing future fate and effect assessment with the aim of establishing a test of chemicals for experimental testing;
4. Developing a methodology for integrating QSARs models and other alternative methods to experimental testing in a probabilistic risk assessment framework, based on characterisation of uncertainty and variability of the models and alternative methods;
5. Dissemination activities like establishing and operationalising of the public CADASTER website, newsletter, and organisation of workshops.

For each of these activities, the main findings are indicated below:

1. Despite the large number of experimental data, in general only limited data are available for the endpoints of primary interest for environmental risk assessment within REACH, i.e. the SIDS endpoints. Especially data on adverse effects for relevant species are lacking and most toxicity data are available for mammals (rats/mice). To supplement the database on SIDS-endpoints of primary interest and to allow for optimal model development and model validation, a limited number of experimental data were generated. The endpoints of interest focused on fate and behaviour in the aquatic environment. The chemicals were selected on the basis of multivariate experimental design supplemented with necessary pragmatic considerations of availability of the chemicals selected.
2. As previously reported, the majority of the existing QSARs are not externally validated, and a definition of the applicability domain is in general lacking. Thus, they do not fulfil the ‘OECD principles for QSAR validation’ for regulatory applicability. Hence, these models are of limited utility for the specific classes of compounds studied under the project CADASTER, and for implementation in REACH in general. The Partners of CADASTER developed new models by applying different modelling approaches and taking into account the OECD principles for validation and acceptability of QSARs for regulation purposes (in particular external validation and assessment of the applicability domain). A comparison between local models developed for aquatic toxicity of (benzo)triazoles by the University of Insubria (UI models) and ECOSAR models (baseline toxicity model and model for Triazoles) highlighted a higher accuracy in prediction (measured by RMSE) of the local UI models. The same was already demonstrated in previous report for phys-chem properties of PBDEs, PFCs, BTAZs, in comparison with EPISUITE predictions.
3. As a follow-up on a previous gap analysis, a set of chemicals with priority of testing was established. The chemicals within the test set were selected by means of principal component analysis using structural information on the chemical classes of interest, and experimental data on toxicity and physic-chemical properties already available for chemicals within the four chemical classes of interest to CADASTER.
4. It is concluded that the characterization of predictive uncertainty would benefit from a probabilistic formulation of QSAR models and models with point estimates as output may be turned into a probabilistic framework without any loss of validity from a chemical point of view. However, a QSAR model for use in probabilistic risk assessment needs to be validated for its ability to make reliable predictions and to quantify associated uncertainty.
5. The QSPR-THESAURUS web site was further updated (amongst others by implementation of the methods developed within CADASTER for assessing the domain of applicability of models, by optimization of 3D structure and by establishing procedures for dealing with different versions of software), a software project was launched to integrate CADASTER models with OpenTox, the offer was made to make data and models collected within CADASTER available for implementation in the OECD QSAR toolbox, the newsletter was established, and the first workshop was organised in order to allow for training of a new generation of chemical risk assessors.

In addition to these main activities, and given one of the aims of REACH of obtaining a proper balance between societal, economic and environmental objectives, a case study was devoted to analyzing the trade-offs between the benefits and the environmental impacts of chemical substances. In general terms, such an analysis is not straightforward as the benefits of the production and application of chemicals can relatively easily be expressed in monetary units, whereas no methods have been established to capture ecosystem properties and ecosystem processes in commercial markets. To obtain objective cost estimates, we developed a model designed to quantify the direct costs that are required to keep the number of animals in a wildlife population on the level corresponding with uncontaminated conditions as a measure of ecosystem impact. The model was applied to assess the costs of the impacts of polybrominated diphenyl ethers (PBDEs) on a population of peregrine falcons, including assessment of the uncertainties in future environmental concentrations of PBDEs by means of implementing three future exposure scenarios.

The REACH regulation advocates the use of non-animal testing methods for risk assessment of chemicals, but guidance is needed on how these methods should be used. The procedures include alternative methods such as chemical and biological read-across, in vitro results, in vivo information on analogues, (Q)SARs and exposure based waiving. The concept of Intelligent Testing Strategies (ITS) for regulatory endpoints has been outlined to facilitate the assessment.

CADASTER aims at providing the practical guidance to integrated environmental risk assessment procedures by carrying out a full environmental hazard and risk assessment for chemicals belonging to four selected compound classes. The QSAR models and data produced during the project will be made publicly available and accessible to all interested partners and stakeholders. They will be available as an on-line and standalone tools and will be compatible for integration within the OECD QSAR Toolbox. It will be freely distributed to SMEs and other industrial partners. The support and development of new models will be continued by the company eADMET GmbH.

Workshop Chairs

Prof. Willie Peijnenburg (RIVM, project coordinator) and Dr. Igor V. Tetko (HMGU, local organizer)

Local Organizing Committee

Dr. Eva Schlosser (HMGU), Stefan Brandmaier (HMGU), Pantelis Sopasakis (HMGU), Dr. Igor V. Tetko (HMGU)

Registration

To register for the workshop, please visit this site and fill in the form until August 31^st 2012. In case you have problems with the registration, you can also send an eMail containing the required information to workshop2012 [at] cadaster.eu. There is no conference fee.

To submit an abstract for poster presentation, please click here

Scientific Programme

There will be two sessions:

• Risk assessment;
• QSAR / QSPR modelling.

The full up-to-date agenda is available at the dedicated workshop webpage.

At present, human health and environmental risk assessments are typically conducted independently of each other. This means that data from toxicological and ecotoxicological studies are not readily accessible by risk assessors in the other discipline, and possibilities for synergies are not realised. Risk assessors in both disciplines are facing challenges due to the increasing need for risk assessments (e.g. REACH, toxicity of mixtures), public and legislative pressure to reduce the amount of animal testing, and budget restrictions. It is clear that better coordination and exploitation of existing data is needed to optimize resource use in human and environmental risk assessment. The objectives of HEROIC include:

• Contribute to the harmonisation of tools and methods in human and environmental risk assessment by exploring how hazard characterisation and exposure assessment data can be used across disciplines;
• Develop a framework for integrated approaches and methodologies for human and environmental risk assessment (for all chemical classes, also including mixtures);
• Facilitate better understanding and co-operation between stakeholders involved in human and environmental risk assessment, in order to improve the quality, perceived value, and acceptance of integrated risk assessment, thereby helping to improve risk management decisions.

The addopted aproaches are:

• Produce a comprehensive map of key stakeholders, risk assessment processes and legal frameworks, to identify current approaches in risk assessment and priorities for future harmonisation and improvement;
• Examine human and environmental data and models available for hazard characterisation and exposure assessment;
• Evaluate how far these test data and models can be used for extrapolation between human and environmental risk assessment (examining species homologies, relevance of endpoints and modes of action);
• Propose an integrated decision-making framework, using a weight-of-evidence approach, and develop novel integrated testing strategies for data-poor situations;
• Communicate the knowledge produced by the project to the relevant audiences, using dedicated newsletters, peer-reviewed journal articles, workshop proceedings, etc. on the purpose-built open access website;
• Set up a dedicated data and training platform “Tox-Hub” on the project website (publicly accessible, also after the conclusion of the project).

The expected impacts include:

• Better coordination and exploitation of resources by establishment of a common framework for human and environmental risk assessment, resulting in optimal use of resources;
• Improved quality of human and environmental risk assessments based on novel integrated testing strategies in data-poor situations;
• Improved capacity building for risk assessors across EU member states, through the dedicated data and training platform “Tox-Hub”;
• Improved stakeholder and public understanding and acceptance of integrated risk assessment by open-access communication of the knowledge produced within the project.

More information is available at the HEROIC website.

New data were generated on endpoints and chemicals for which insufficient data were available for model validation and proper hazard/risk assessment. Following evaluation of available experimental data and available (Q)SAR models, new experimental data on the most essential endpoints of assessment (i.e. new data for the endpoints that are responsible for most of the variance in the safety, hazard and risk assessment) were generated.

The main topic of study were emerging compounds that might develop into the future chemicals of concern for which typically limited data are available. Based on an initial evaluation of the availability of experimental data and QSAR models for various classes of the emerging compounds identified within NORMAN, four groups of compounds have been selected as the chemical classes of choice for CADASTER. The following toxicity, fate and behaviour testing was performed on the selected classes:

1. Polybrominated diphenylethers (PBDE): 28-day sediment testing of PBDEs was performed on bioaccumulation with aquatic oligochaeta Tubifex tubifex by PHI;
2. Perfluoroalkylated substances and their transformation products: Toxicity testing of fluorinated compounds was performed with lettuce (Lactuca sativa) and green algae (Pseudokirchneriella subcapitata) at RIVM. Thereupon, testing was performed with two cladoceran species (Daphnia magna and Chydorus sphaericus), as well as with embryos of the zebrafish (Danio rerio), also at RIVM;
3. Substituted musks/fragrances: Toxicity testing of fragrances was performed with green algae (Pseudokirchneriella subcapitata) and with Daphnia magna at PHI. Substituted musks/fragrances were tested also on ready biodegradability, at PHI;
4. Triazoles/benzotriazoles: Toxicity testing of substituted (benzo)triazoles was performed with Daphnia magna and with embryos of the zebrafish (Danio rerio) at RIVM. Toxicity testing of substituted (benzo)triazoles was performed with green algae (Pseudokirchneriella subcapitata) at PHI. Substituted (benzo)triazoles were tested also on ready biodegradability, at PHI.

Report on the newly data generated is available here.

Some new models for aquatic and terrestrial toxicity of (B)TAZs have been developed and presented during the last year at SETAC conferences in USA (32^nd Annual Meeting SETAC North America, November 2011) and Europe (6^th SETAC World Congress, Berlin 2012). WP3 partners (UI, LNU, IVL, IDEA and HMGU) developed different QSAR models for the prediction of aquatic toxicity of (B)TAZs by applying different modelling approaches (e.g. MLR-OLS, PLSR, Kohonen Neural Network) starting from theoretical molecular descriptors calculated by commercial and freely available software (DRAGON, PaDEL-Descriptor, CADASTER web). The studied end-points were EC50 (72h) in Pseudokirchneriella subcapitata, EC50 (48h) in Daphnia magna, and LC50 (96h) in Onchorhynchus mykiss, since acute toxicity data for algae, zooplankton and fish are the basic information required for risk assessment of chemicals. QSARs were checked for their goodness-of-fit, robustness, predictivity and applicability domain, in agreement with OECD principles for the validation of QSARs for regulatory purposes. Models were also applied for the screening of more than 300 (B)TAZs without experimental data, many of them included in the European Chemicals Agency (ECHA) list of pre-registered substances and therefore to be registered under REACH. This screening allowed for identification of the most problematic compounds for the aquatic environment. A comparison with ECOSAR predictions showed a higher prediction accuracy of the local models developed within the CADASTER project. Finally, consensus predictions obtained by combining different WP3 models will be evaluated using the experimental data measured by W2 Partners (PHI and RIVM) for algae and daphnia acute toxicity.

Terrestrial toxicity of (B)TAZs was also investigated by UI, which developed ad hoc classification models for the prediction of acute toxicity on earthworm, honeybee and bird, three key organisms commonly considered for the terrestrial risk assessment. Also in this case, the proposed models fulfill the OECD requirements for QSAR acceptability in regulation (i.e. internal and external validation, a defined applicability domain and interpretation of modelling descriptors).

QSAR classification models for the prediction of ready biodegradability of fragrances are now under development by WP3 Partners (UI, IDEA and HMGU).

A CADASTER report “Application of QSAR models for probabilistic risk assessment” has recently been accomplished. This report is a deliverable within the CADASTER project with an aim to show how to increase the use of non-testing information for regulatory decision whilst meeting the main challenge of quantifying and reducing uncertainty. The report contains examples of the integration of testing and non-testing information into assessment models for carrying out safety-, hazard- and risk assessments. The application of QSAR models for probabilistic risk assessment was discussed in respect to the characterization of uncertainty in QSAR predictions, the propagation of uncertainty in the assessment and sensitivity analysis of individual QSAR predictions with regard to their contributions in the overall risk assessment framework. Integration of QSARs in chemical safety assessment must acknowledge that treatment of uncertainty is context dependent, and uncertainty is to be interpreted in relation to the kind and quality of background information.
Two case-studies were used to demonstrate the computational framework for QSAR based risk assessment. The lesson learnt was that the application of QSARs in probabilistic risk assessment leads to questions such as:

• Are there any QSAR data available to use as weight-of-evidence of a chemico-specific input parameter to the exposure and effect assessments?;
• Which validated algorithm for prediction and approach for predictive inference will be used to support the uncertainty analysis?
• Is a QSAR prediction reliable enough to support the intended decision making?

The first case-study was an uncertainty and sensitivity analysis of QSAR and QSPR based-risk assessment on five Triazoles. Soil sorption partition coefficients, solubility, melting point, vapor pressure and hydroxyl radical reaction in air were predicted with identified QSPRs, and no effect concentrations with available QSARs developed for Triazoles. Parameter uncertainty was treated as a probability distribution for the chosen QSAR algorithm, for example a t-distribution when the QSAR algorithm was an ordinary least square regression. Exposure and effect assessment were carried out in the multimedia fate model Simplebox, and uncertainty was propagated by Monte Carlo simulation. Uncertainty analysis showed that the maximum permissible emissions to agricultural soil were highest for the Triazoles Bromuconazole and Difenoconazole. Sensitivity analysis, aimed to find sources of uncertainty with a large influence on the uncertainty in assessment output, found that uncertainty in maximum permissible emissions to agricultural soil was mainly determined by uncertainty in the soil sorption partition coefficient, the biodegradation in water and the toxicity.

The second case-study compared the discrepancy between non-testing versus testing-based risk assessments on three polybrominated diphenyl ethers. Exposure (PEC) and effect (PNEC) were assessed based on QSPR and QSAR predictions, treated in the same way is the first case-study except that variability in effect was considered through a species sensitivity distribution. Whenever available, parameters for the exposure assessment and species effect concentrations were replaced by information from experimentally tests on the chemicals in question. Discrepancies between non-testing and testing-based risk assessments were evaluated by studying the range of emissions for which the resulting decisions were different. We found that non-testing versus testing based risk assessments were different, but this difference depend to a large extent to how uncertainty is dealt with. The difference is small when predictions are precise, i.e. providing a good but balanced coverage of the experimental values, non-testing information are useful complement to reduce uncertainty in existing testing information of effects. However, there is uncertainty in experimental data which are not accounted for in QSAR modeling. The larges discrepancies could however be related to the use of uncertainty factors. For example, PNEC values were given a different treatment of uncertainty depending on the number of species with available information, where QSAR predictions increased the number of covered species. Also, experimental information was available for acute effects, whereas QSAR predictions were available for chronic effects, was another cause for difference in the treatment of uncertainty.

With respect to treatment uncertainty in QSAR predictions for probabilistic risk assessment the conclusions in the report are that:

1. The integration of QSARs into probabilistic risk assessment is possible given proper assessments of predictive uncertainty and predictive reliability. Predictive uncertainty and reliability are identified to inform the characterization of parameter and model uncertainty, two kinds of uncertainty to be identified in probabilistic risk assessment. Uncertainty in probabilistic chemical safety assessment can roughly be divided in to three categories: parameter uncertainty, model uncertainty and scenario uncertainty. According to ECHA scenario uncertainty is “the uncertainty in specifying the scenario(s) which is consistent with the identified use(s) of the substance”. Scenario uncertainty is of less relevance for the uncertainty related to a QSAR. The other two terms, parameter and model uncertainty, are not frequently used in QSAR modeling. Instead a distinction lies between predictive uncertainty and predictive reliability.
2. Probabilistic risk assessment is supported by QSAR predictions derived from Bayesian predictive inference. Predicting must be done with care, and the use of different bases for predictive inference is possible when a QSAR is treated as a scientific based hypothesis supported by empirical data. In other words, the suggestion is to treat QSAR data that has been validated for its predictive performance as the QSAR to base predictions on. It is up to the assessor to choose an appropriate method for predictive inference, given that performance measures of predictivity do not deviate to much from the measures in the peer-reviewed validation. This will result in a more flexible use of QSARs with a possibility of updating as long as new QSAR data becomes available.
3. The extent of extrapolation in a QSAR prediction influences predictive error and predictive reliability, and the domain of applicability is from an applied perspective context dependent and considered in the treatment of uncertainty. A separation between predictive uncertainty and predictive reliability makes it possible to both apply models and discuss their reliability in a constructive way.

This report focuses on predictions of chemical specific properties and activities to replace testing information in chemical regulation. Statistical inference may have different purposes, and when QSARs are applied to support decisions making based on unobserved quantities such as in risk assessment, drug development, or experimental design, the statistical problem is to make predictions, referred to as predictive inference. Predictive science involves the use of a belief system about observables in science, and a philosophy of scientific methodology that implements that belief system. Predicting should be done with care and there are (solvable) practical problems when the purpose of statistical inference changes from inference on models to inference on predictions.

The report discusses three kinds of philosophies for predictive inference of relevance for the application of QSARs in probabilistic risk assessment. Sampling Theory estimate predictive uncertainty based on a representative sample. Such (frequentist) inference rests upon assumptions of independent and, for example, identically distributed observations, in combination with a probabilistic assumption of uncertainty. Under violence of any of these assumptions, appliers of frequentist inference run into problems.

The Bayesian paradigm for inference assign, instead of assume, a probabilistic model for observations, and assign models for uncertainty in parameters (so called priors). Bayesian inference uses Bayes rule to update expert knowledge with information in empirical observations. The result is a well-defined probabilistic model of uncertainty. In cases of doubts, the caveat is the necessity to choose priors and probabilistic models (likelihoods). For example, there is no need to check an assumption of normality of errors (as in the frequentist case), as this is assigned through expert judgment.

The third alternative is to assign a probability distribution for predictive uncertainty based on expert judgment only. This can for example, be based on experience of experimental testing, or based on combinations of different sources of information.

Sampling Theory and solid expert judgment can be seen as extremes kinds of Bayesian inference; the first as Bayesian inference with non-informative priors (expressed simplistically, but it is more difficult than that); the second as Bayesian inferences with only priors. Therefore the recommendation we give is to use Bayesian inference as the statistical philosophy for predictive inference when QSARs are applied in probabilistic risk assessment.

In our first study, we showed that a stepwise refinement of the representation of the chemical space can be used to significantly decrease the number of required experiments. This refinement is implemented by taking the growing amount of information, which is gathering within the experimental testing cycles, into consideration. The correlation between the measured property and the descriptor space is thereby the criterion for the rearrangement of the search space. To evaluate the newly developed idea, we compared the performance of models derived with the D-Optimal criterion such a representation of the chemical space to that of models derived using the same selection criterion, but applied in a static manner on principal components. The validation was done on four datasets with different endpoints, representing toxicity, physicochemical properties and an adsorption coefficient. Compared to models derived from the selection on principal components, the models derived from the selection on PLS latent variables had a lower RMSE and a higher Q2 and an increased correlation coefficient. Further, we were able to produce models of the same quality with testing only 60% of compounds.

As a stepwise selection process is not feasible for all problems, we additionally investigated on the most efficient non-stepwise selection criterion. We compared approaches like the classical D-Optimal criterion, the Kennard-Stone algorithm, full factorial design, a representativity selection to a selection based on the k-Medoid clustering. The results show that the performance of models developed on selections based on the k-Medoid approach is among the best for all examined datasets. It is the only approach to deliver significantly better results on all datasets than a random selection. Compared to the models derived from the compounds selected by the other examined approaches, those derived with the k-Medoid selection show a high reliability for experimental design, as their performance was constantly among the best for all examined datasets. The reasons for this exceptional performance are:

• the whole chemical space of relevance is covered;
• each compound in the periphery of the data cloud is represented in a cluster;
• from each cluster the most representative compound is selected.

A third study to evaluate experimental design strategies in terms stability, reliability and the robustness against structural outliers is currently in progress.

The workshop involved 35 participants, including invited speakers from JRC - Institute for Health & Consumer Protection (Italy), ECHA – Evaluation Unit (Finland), Douglas Connect (Switzerland) and University of North Carolina USA).

On September 1^st there were presentations of CADASTER partners giving an overview of the CADASTER project and the results of the project:

• General philosophy of CADASTER (Willie Peijnenburg RIVM, The Netherlands);
• CADASTER achievements (Mojca Kos Durjava, PHI Maribor, Slovenia).

In addition three invited speakers gave an overview on the FP7 OpenTox project and on different aspects of alternative methods in REACH:

• How OpenTox satisfies REACH requirements (Barry Hardy, Douglas Connect, Switzerland);
• Legislative overview on the use of alternative methods in REACH (Evelin Fabjan, ECHA, Finland);
• Technical information on alternative methods (Andrew Worth, JRC, Italy).

On September 2^nd, the fourth invited speaker gave an overview about the state of the art on QSARs and CADASTER partners gave presentations about the CADASTER achievements:

• Information about state of the art on QSARs (Alex Tropsha, University of North Carolina, USA);
• CADASTER achievements (Paola Gramatica, University of Insubria, Italy);
• CADASTER results (Tomas Oberg, Linnaeus University, Sweden; Igor Tetko, HMGU, Germany).

In addition, on-line tools training sessions were organised:

• Training lessons for on-line tools that can be used to estimate REACH end-points for chemical compounds and decrease the number of animal tests: Demonstration of the tools developed in CADASTER (Igor Tetko, HMGU, Germany);
• Demonstration of the models available in the OECD QSAR toolbox (Emil Rorije, RIVM, The Netherlands).

The workshop was closed with a panel discussion on the use of QSARs in REACH chaired by Willie Peijnenburg (RIVM, The Netherlands).

The subsequent feed-back of the workshop participants on their experiences with the implementation of REACH was of importance for the customization of the tools for the use by the regulatory agencies. A report on the workshop is available here.

Three posters and one spotlight presentation were presented in the session on “Quantitative Structure Activity Relationship (QSARs) and similar models for predicting the toxicity of chemicals, mixtures and combined stress”. In particular:

• one poster, presented by Stefano Cassani, summarized the QSAR models developed by all WP3 partners for the prediction of aquatic toxicity (algae, daphnia and fish) of triazoles and benzotriazoles;
• one spotlight presentation and one poster were presented by Simona Kovarich, and were focused on the classification models developed by UI for the prediction of (B)TAZs toxicity toward three key organisms of the terrestrial ecosysthem (earthworms, honeybees and birds);
• one poster was presented by Paola Gramatica, and it was focused on the comparison of various validation parameters and the new thresholds proposed by UI for the evaluation of the external predictivity of QSAR models.

Additionally, the poster “Prioritization of emerging pollutants on the basis of chemical structure” was presented by Simona Kovarich in the session “Non-target analysis and identification of toxicologically significant emerging pollutants”. This poster, summarized the prioritization activity applied to CADASTER chemicals (BFRs, fragrances, PFCs and (B)TAZs) in order to focus the experimental design on critical substances on the basis of their chemical structure and of their potential ecotoxicological hazard. The prioritization study was performed to identify the priority compounds that were suggested for focusing the experiments executed by other CADASTER partners (WP2). This poster generated the highest interest among SETAC participants. This fact highlights the importance of in silico chemical screening finalized to reduction of experimental tests, especially in the context of REACH.

Besides WP3 partner UI, WP4 partner RU also participated in the meeting. In the session “Extrapolation within wildlife toxicology”, Laura Golsteijn held a presentation entitled “Do interspecies correlation estimations increase the reliability of toxicity estimates for wildlife?”. She described a study in which an experimental dataset used for calculation of the hazardous dose for warm-blooded species was enlarged with interspecies correlation estimates. It was found that small experimental datasets often underestimate the hazardous dose of a chemical, and that combined use of lab data and model predictions reduces high statistical uncertainty. The findings illustrate again the need for integration of experimental information, predictive models and strategies for carrying out safety-, hazard- and risk assessments for large numbers of substances.

15th International Workshop on Quantitative Structure-Activity Relationships in Environmental and Health Sciences,18-22 June 2012, Tallinn, Estonia

Participants: Nina Jeliazkova (Ideaconsult) Paola Gramatica (University of Insubria) Ester Papa (University of Insubria) Igor V. Tetko (Helmholtz-Zentrum München)

Presentations: Nina Jeliazkova Linking the silos: Data and predictive models integration in toxicology (Plenary lecture) Paola Gramatica Externally Predictive QSAR Models: Thresholds of Acceptance by Various External Validation Criteria and Critical Inspection of Scatter Plots (Plenary lecture) Ester Papa, P. P. Roy, S. Kovarich, S. Cassani, L. Van der Wal, P. Gramatica Terrestrial Toxicity of (Benzo) Triazoles Modeled by QSAR (Poster) Ester Papa, S. Cassani, S. Kovarich, P. P. Roy, M. Rahmberg, S. Nilsson, N. Jeliazkova, N. Kochev, O. Pukalov, I.V. Tetko, P. Gramatica QSAR Models for Aquatic Toxicity of Triazoles and Benzotriazoles: WP3 Results within the CADASTER Framework (Poster) Ester Papa, S. Kovarich, J. Li, P. Gramatica In Silico Analysis of the Endocrine Activity of PFCs and BFRs (Poster) Igor V. Tetko, Iurii Sushko, Stefan Brandmaier, Paola Gramatica, Ester Papa, Simona Kovarich, Barun Bhhatarai, Mojca Kos Durjava, Nina Jeliazkova, Ullrika Sahlin, Magnus Rahmberg, Willie J.G.M. Peijnenburg CADASTER QSPR THESAURUS database (Poster) Stefan Brandmaier, Tomas Öberg, Igor V. Tetko The application of the k-Medoid clustering for representative subset selection in QSAR modeling (Poster) Stefan Brandmaier, Igor V. Tetko Performance, Reliability and Robustness - A comparison of several experimental design strategies (Poster)

Gordon Research Conference – Green Chemistry,22-27 July 2012, Lucca, Italy
Participants: Paola Gramatica (University of Insubria)	Presentations: E. Papa, S. Kovarich, L. Ceriani, S. Cassani, R. Boethling, Paola Gramatica QSAR Study of Fragrances Biodegradability for Greener Chemicals (Poster)

19th EuroQSAR: Knowledge Enabled Ligand Design,26-30 August 2012, Vienna, Austria
Participants: Nina Jeliazkova (Ideaconsult)	Presentations: Nina Jeliazkova, Ola Spjuth, Egon Willighagen Bioclipse-OpenTox: Interactive predictive toxicology (Poster)

XIII Congresso Nazionale di Chimica dell’Ambiente e dei Beni Culturali (Italian Chemical Society), 10-14 September 2012, Taranto, Italy
Participants: Simona Kovarich (University of Insubria) Stefano Cassani (University of Insubria)	Presentations: Simona Kovarich Ready biodegradation QSAR models for fragrances (Oral presentation) Stefano Cassano QSAR models for (B)TAZs aquatic toxicity (Poster)

33rd Annual Meeting SETAC North America, 10-15 November 2012, Long Beach, California, USA
Participants:	Presentations: Ready biodegradation QSAR models for fragrances (Poster)