The work was executed via. 6 work packages. In the first work package (WP1) the two key tasks were executed: 1) Literature survey and collection of the available experimental data representing acute and chronic aquatic toxicity, biodegradation, and bioaccumulation, 2) Chemical and biological data curation of the collected data.
In the second work package, the first task was initially planned to develop new descriptors specific for studying salts, organo-metallic, inorganics, and mixture. However, based on the literature survey we realized that much of the work is already going on with the development of descriptors for the mentioned types of chemicals, and thus we planned to perform a study to evaluate the applicability and efficiency of typically used descriptors as well as new descriptors recommended for handling such chemicals. Thus, as a case study, we have developed a generalized nanoQSAR model for predicting cytotoxicity and genotoxicity of metal oxides nanoparticles (published here DOI: 10.4018/IJQSPR.20201001.oa2). The next task was the development of the QSAR methodology for mixtures. There were two key areas where we can contribute to this topic, i.e. first to compile the information and design a rational methodology, and second to move one step ahead and develop a user-friendly software platform ‘ProtoML-Mixture’ to execute the designed methodology for handling mixtures. To demonstrate the capabilities of the developed software ‘ProtoML-Mixture’ we have performed a case study, where we have developed a multi-tasking QSTR model for predicting the ecotoxicity of deep eutectic solvents (DESs).
In the third work package, we have developed an artificial intelligence (AI) based 'Proto-ML' software to develop and validate the regression and classification-based QSAR models. It has a lot of functionalities that will help users to develop robust models and it works like a workflow, which makes it highly user-friendly. It comprises several essential nodes starting from reading and confirming the input data, data pre-treatment, data set splitting, machine learning (ML) technique node comprising several feature selection methods, ML techniques, as well as, options to set validation parameters, graphs/plots, and applicability domain determination method and finally output data node. In the fourth work package, we have developed several QSTR/QSAR models including multi-tasking that assisted us to screen potential pharmaceuticals and/or cosmetics that may show aquatic toxicity. Thus, first, we have developed requisite QSTR models to predict the acute and chronic toxicity of pharmaceuticals and cosmetics, as well as, to predict the biodegradation and bioaccumulation status. Later, these models were ultimately utilized in screening marketed pharmaceuticals and cosmetics to identify potential chemicals that can be harmful to aquatic life. On screening 8282 chemicals, we have selected 55 chemicals from different toxicity categories for experimental validation of the developed models. The toxicity studies on Daphnia magna were performed in Xenobiotics S.L. (secondment 1), while studies on Oncorhynchus mykiss were performed in INIA, Madrid (secondment 2). We have disseminated the progress and results of the project in about 10 scientific events including conferences, workshops, etc., and promoted in social media.
