Search for: All records

Polymeric coatings are extremely important and widely applied materials. However, the widespread usage of polymeric coatings has increased the likelihood of ecosystem pollution, especially in aquatic environments, posing toxicity risks for microorganisms. Using in silico approaches is a cost- and time-efficient way to assess properties of materials and to reduce potential environmental impact. In this work, a dataset of polymeric materials is collected that includes information on the toxicity of green algae towards different polymeric materials and their particles. A machine learning-based cheminformatics technique known as Quantitative Structure–Activity Relationship (QSAR) was applied to establish a correlation between structure of polymeric particles and toxicity. In result, a predictive model was developed to predict the green algae growth inhibition (EC50). A set of linear and non-linear machine learning (ML) methods was applied to build the structure-toxicity relationship model. Special mixture-based descriptors were generated to apply in QSAR modeling to describe complex polymeric materials and identify the high-relevance features/descriptors responsible for growth inhibition. The best models were selected based on their performance on training (R2train, Q2, RMSE) and validation sets (R2test, y-scrambling). A six-variable model was selected as the best one with R2train = 0.91 and R2test = 0.82, for training and test sets, respectively. In addition, the developed QSAR model identified several influential descriptors, such as hydrogen bond donor fragments, spatial arrangement, and hydrophilicity of the polymeric materials investigated that were responsible for the toxicity.