Quasi-SMILES as a tool to utilize eclectic data for predicting the behavior of nanomaterials

• Quasi-SMILES for representation of conditions (not structure) are suggested.
• Predictive models calculated with quasi-SMILES are suggested.
• The statistical quality of these models is quite good.

Nowadays, nanomaterials are often considered a scientific hit. However, despite the immense advantages of nanomaterials, there are studies, which have shown that these materials can also harmfully impact both human health and the environment. A preliminary evaluation of the hazards related to nanomaterials can be performed using predictive models. The aim of the present study is building up a single QSAR model for predicting cytotoxicity of metal oxide nanoparticles on (i) Escherichia coli (E. coli) and (ii) human keratinocyte cell line (HaCaT) based on the representation of the available eclectic data, encoded into quasi-SMILES. Quasi-SMILES are an analog and an attractive alternative of traditional simplified molecular input-line entry systems (SMILES). In contrast to traditional SMILES quasi-SMILES are a tool to represent not only molecular structures, but also different conditions, such as physicochemical properties and experimental conditions. The statistical quality of the models is average correlation coefficient (r2) and root mean squared error (RMSE) for the training set 0.79 and 0.216; the average r2 and RMSE for validation set are 0.90 and 0.247, respectively.