Pólis

This tool allows for the mutagenicity assessment of chemical compounds based on a consensus strategy: Different chemical representations (molecular fingerprints, 2D and 3D descriptors and molecular graphs) and modelling methods, i.e., Quantitative Structure−Activity Relationship (QSAR), read-across and deep learning models are integrated. The models involved in the consensus scheme are a XGBoost model based on molecular descriptors (predictive accuracy in external validation 81%), a graph convolutional neural networks model (predictive accuracy in external validation 81%) and a read-across methodology based on a guided-k-Nearest Neighbours scheme (guided-kNN) where two different molecular representations (molecular fingerprints and descriptors) are considered for neighbour selection and predictions generation (predictive accuracy in external validation 80%). The mutagenicity predictions from the three models are integrated in a majority voting scheme to enhance the overall predictive accuracy (83% in external validation) and reducing individual model biases.

Within this web-tool, users can import compounds of interest by either drawing molecules using an integrated sketcher or uploading a list of molecules in SMILES format. Users can also inspect a 3D visualization of their molecular structures post-drawing. The platform can be used for high-throughput virtual screening through batch processing: users can upload SDF files and process multiple compounds with one query. The results include for each input structure the predicted mutagenicity class (mutagen/non-mutagen) and an indication of its reliability (High/Moderate/Low). All predictions can be downloaded per model for further analysis, enhancing flexibility and usability for diverse research applications. Apart from the predictions for the XGBoost and the guided-kNN model their reliability according to the corresponding applicability domain is presented. Especially for the developed guided-kNN models, in the downloaded results the training neighbours and their similarity from the input substances are presented. The dataset used for model development can be found in the ChemPharos DB