Virtual chemical analysis and machine learning-based prediction of polyethylene terephthalate nanoplastics toxicity on aquatic organisms as influenced by particle size and properties

This study focuses on the chemical analysis and prediction of Polyethylene Terephthalate (PET) nanoplastics toxicity on aquatic organisms, considering the influence of particle size and properties. The effect PET NPs of different sizes (1, 4, 9, 16 and 25 nm coded NP1 to NP5) on aquatic organisms such as Terpedo californica (electric ray fish) and Danio rerio (zebrafish) as model species was evaluated by virtual chemical techniques and machine learning methodology based on Multilayer Perceptrons Artificial Neural Networks (MLP ANN) and Support Vector Machine. The PET NPs was built and characterized in silico and then docked on the acetylcholinesterase (TcAChE) and cytochrome P450 (Zf CYP450) of the organisms, respectively. The results showed that the binding affinities of the NPs increased steadily from – 7.1 kcal mol-1 to – 9.9 kcal mol-1 for NP1 to NP4 and experienced a drop at NP5 (– 8.9 kcal mol-1) for TcAChE. The Zf CYP450 also had a similar pattern ranging from -5.2 kcal mol-1 to -8.1 kcal mol-1. The MLP ANN showed an accuracy of 85.9 % and 77.3 %. In comparison, SVM showed a better PET NPs toxicity prediction with an accuracy of 99.5 % and 99.4% based on the inherent properties of TcAChE and Zf CYP450, respectively.