Junli Xu1,2,3
1School of Biosystems and Food Engineering, University College of Dublin, Belfield, Dublin 4, Ireland.
2Institute of Food and Health, University College Dublin, Belfield, Dublin 4, Ireland.
3Conway Institute, University College Dublin, Belfield, Dublin 4, Ireland.
junli.xu [at] ucd.ie
Abstract
Numerous articles have been published investigating the health effects of exposure to micro- and nanoplastics (MNPs). However, these studies have yielded inconclusive findings due to the lack of comparability between them and the complex and diverse nature of the existing toxicity data on MNPs. This study presents a predictive modeling framework for assessing the cytotoxicity of MNPs using machine learning techniques based on classification. Through a thorough literature search, a dataset comprising 1824 sample points was compiled, incorporating nine features that describe the physicochemical properties of MNPs, cell-related attributes, and experimental factors. The decision tree ensemble classifier constructed using all the features (referred to as DTE1) exhibited a high predictive accuracy of 0.95, along with a recall and precision of 0.86 each. To identify the key factors influencing the toxic properties of MNPs, feature selection was performed. A simplified classifier utilizing six influential features demonstrated a comparable performance to DTE1. These findings can guide future studies by improving experimental design and reporting practices, ultimately enhancing our understanding of the urgent health concerns related to MNPs. As more representative research data is incorporated, the developed model holds the potential for broad applicability in various settings concerning MNP cytotoxicity.
Keywords: microplastic; nanoplastic; cytotoxicity; health effect; machine learning
Acknowledgement: Funding for this research was provided by the Science Foundation Ireland (SFI)-Irish Research Council Pathway Programme Proposal ID 21/PATH-S/9290.
