QSAR prediction of aquatic and mammalian toxicity of triazoles and benzo-triazoles

Triazoles and benzo triazoles (TAZ/BTAZ) are potentially hazardous chemicals that adversely affect humans and other non-target species, and are on the list of substances of very high concern (SVHC) in the European regulation of chemicals REACH. TAZ/BTAZ are synthetic molecules used in various industrial processes (to obtain pharmaceuticals and agricultural products), and have a wide application as anti-corrosives, cleansing agents for textiles, flame retardants, photographic emulsions, etc…Furthermore they are abundantly used as components of liquid deicing agents for aircraft and airport runways. Because of their wide use they have been found distributed throughout the environment, mainly in water compartments. The amount of experimental data available for these molecules is insufficient for a comprehensive characterization of their environmental and toxicological profile and they have been included among the four classes of chemicals studied in the European FP7 Project CADASTER (CAse studies on the Development and Application of in Silico Techniques for Environmental hazard and Risk assessment). In this study we investigated and modeled by QSAR different endpoints of interest to define the potential aquatic toxicological profile of hundreds of TAZ/BTAZ, and the possible correlations among their aquatic and mammalian toxicity. The studied end-points were: LC50 in Onchorhynchus Mykiss, EC50 in Daphnia Magna, and EC50 in algae. Data for mammalian acute toxicity in rat (LD50 oral exposure) were also investigated and modeled. Different theoretical molecular descriptors were calculated by different proprietary and freely available online software (DRAGON, Hyperchem, and the CADASTER online platform for the calculation of molecular descriptors – www.cadaster.eu). The endpoints of interest were modeled by multiple linear regression (MLR) and the Genetic Algorithm was used to select the relevant molecular descriptors by the MLR-Ordinary Least Squares (OLS) method. The best models were validated for their predictive performance using leave-one-out, bootstrap and the scrambling of the responses. External validation was also performed depending on the dimension of the studied experimental datasets. The reliability of the predictions was always checked by the leverage approach in order to verify the chemical applicability domain of the models.