Evaluation of CADASTER QSAR models for the aquatic toxicity of (benzo)triazoles and prioritisation by consensus prediction.

QSAR regression models of the toxicity of triazoles and benzotriazoles ([B]TAZs) to an alga (Pseudokirchneriella subcapitata), Daphnia magna and a fish (Onchorhynchus mykiss), were developed by five partners in the FP7-EU Project, CADASTER. The models were developed by different methods - Ordinary Least Squares (OLS), Partial Least Squares (PLS), Bayesian regularised regression and Associative Neural Network (ASNN) - by using various molecular descriptors (DRAGON, PaDEL-Descriptor and QSPR-THESAURUS web). In addition, different procedures were used for variable selection, validation and applicability domain inspection. The predictions of the models developed, as well as those obtained in a consensus approach by averaging the data predicted from each model, were compared with the results of experimental tests that were performed by two CADASTER partners. The individual and consensus models were able to correctly predict the toxicity classes of the chemicals tested in the CADASTER project, confirming the utility of the QSAR approach. The models were also used for the prediction of aquatic toxicity of over 300 (B)TAZs, many of which are included in the REACH pre-registration list, and were without experimental data. This highlights the importance of QSAR models for the screening and prioritisation of untested chemicals, in order to reduce and focus experimental testing.

Ambit-GCM: An Open-source Software Tool for Group Contribution Modelling.

Ambit-GCM is a new software tool for group contribution modelling (GCM), developed as a part of the chemoinformatics platform AMBIT. It is an open-source tool distributed under LGPL license, written in Java and based on the Chemistry Development Kit. Ambit-GCM provides an environment for creating models of molecular properties using additive schemes of zero, first or second orders. Ambit-GCM supports a set of local atomic attributes used for dynamic configuration of desired atom descriptions, which are applied to define fragments of different sizes. All defined groups are exhaustively generated for each molecule from a training set of compounds and combined to form the basic set of GCM fragments. Additionally, Ambit-GCM users can define correction factors via custom SMARTS notations or add externally calculated molecular descriptors. A molecular property model is obtained as a sum over all found groups by multiplying each group or correction factor frequency to its corresponding contribution. Multiple linear regression analysis (MLRA) is used for group contributions calculation. Ambit-GCM performs full statistical characterization of the obtained MLRA models via various validation techniques: external tests validation, cross validation, y-scrambling, etc. The software can be optionally used only for molecule fragmentation combined with an external statistical modelling package for further processing. Ambit-GCM example usage and test cases are given.

CADASTER QSPR Models for Predictions of Melting and Boiling Points of Perfluorinated Chemicals.

Quantitative structure property relationship (QSPR) studies on per- and polyfluorinated chemicals (PFCs) on melting point (MP) and boiling point (BP) are presented. The training and prediction chemicals used for developing and validating the models were selected from Syracuse PhysProp database and literatures. The available experimental data sets were split in two different ways: a) random selection on response value, and b) structural similarity verified by self-organizing-map (SOM), in order to propose reliable predictive models, developed only on the training sets and externally verified on the prediction sets. Individual linear and non-linear approaches based models developed by different CADASTER partners on 0D-2D Dragon descriptors, E-state descriptors and fragment based descriptors as well as consensus model and their predictions are presented. In addition, the predictive performance of the developed models was verified on a blind external validation set (EV-set) prepared using PERFORCE database on 15 MP and 25 BP data respectively. This database contains only long chain perfluoro-alkylated chemicals, particularly monitored by regulatory agencies like US-EPA and EU-REACH. QSPR models with internal and external validation on two different external prediction/validation sets and study of applicability-domain highlighting the robustness and high accuracy of the models are discussed. Finally, MPs for additional 303 PFCs and BPs for 271 PFCs were predicted for which experimental measurements are unknown.