Cytotoxicity analysis of nanoparticles by association rule mining

Understanding the toxicity behavior of NPs is of great importance to ensure efficient delivery to intracellular targets without causing cytotoxicity, to measure the long-term effects of nanoparticles (NPs), and to predict risks and hazards to humans and the environment. However, the diversity and complexity of toxicity sample space, individual sampling, and missing/unreported information make it burdensome to draw general conclusions from the whole picture. To overcome the challenges of safe-by-design of NPs, here we create a machine learning approach by knowledge extraction from the literature which is based on the published toxicity data of inorganic, organic and carbon-based NPs. By integrating 15 qualitative and 10 quantitative attributes from 152 articles, we have obtained 4111 instances reflecting the toxicity of NPs with various material, cell, and experimental properties. Implementing association rule mining (ARM), we present that cytotoxicity is primarily related to the core and coating material of NPs, their synthesis route and the exposed cell type. Our model addresses the interdependence of attributes by shedding light on hidden relationships and we claim that it may contribute to the controlled and safe design of NPs in future studies.

Automated summary

Understanding the toxicity behavior of nanoparticles is crucial for efficient delivery to intracellular targets, measuring long-term effects, and predicting risks. A machine learning approach was created based on published toxicity data, integrating various attributes to shed light on associations between nanoparticle toxicity, core material, synthesis route, and cell type. The model addresses interdependencies of attributes and may contribute to controlled and safe nanoparticle design.