Use of dissociation degree in lysosomes to predict metal oxide nanoparticle toxicity in immune cells: Machine learning boosts nano-safety assessment

Abstract

Potential immune responses resulting from exposure to metal oxide nanoparticles (MeONPs) have been the subject of intensive discussion in the last decade. Despite the extensive use of MeONPs in several applications, their toxic effects on immune cells have rarely been predicted in silico because of the complexity of immune responses and the complicated properties of MeONPs. In the present study, machine learning (ML) methods coupled with high-throughput in vitro bioassays were used to develop models for predicting the toxicity of MeONPs in immune cells. An ML model with a high prediction accuracy (97% and 96% in the training and test sets, respectively) was constructed by resolving the class imbalance problem in training and applying an ensembled algorithm. Further, to verify the model, MeONPs outside the scope of the datasets were selected to examine their cytotoxicity experimentally. The model was validated against independent MeONPs, with an accuracy of 91%. ML methods coupled with intracellular imaging revealed that the toxic ions released in the lysosome were an important determinant of toxicity in immune cells. Furthermore, ζ-potential, electronegativity, and size are crucial factors for predicting nanotoxicity. We believe the established modeling framework will provide useful insights for designing and applying safe nanoparticles and facilitating decision-making for environmental and health protection.

Keywords

• Metal oxide nanoparticles
• Machine learning
• Cytotoxicity
• Immune response
• Intracellular dissociation
• Lysosome