Detection of Nanoplastics Within Complex Environmental and Food Resources Matrices Via Machine Learning

The pervasiveness of nanoplastics within the environment underscores the need for robust and accurate methods for their identification and classification. The lightweight and small nanoplastics (NPs) can bypass biological barriers and disperse throughout the environment, posing significant health risks to humans and aquatic life. Typical detection of nanoplastics has relied on cumbersome filtration, and subsequent coloration of the plastics for visualization, once they have been painstakingly separated from their matrix, including fish, sand, and soil using water. Raman spectroscopy, however, offers an alternative, as it effectively detects these particles without the need for separation, with its high resolution (<1µm). Unfortunately, accurate identification and classification are challenging because of the faint Raman scatter of NPs and signal interference from background noise. To address this challenge, this project proposes a method integrating machine learning (ML) with Raman spectroscopy. Multiple ML models were first trained with Raman spectra of 50μg/mL suspensions of PE, PTFE, PS, PMMA, and PVC NPs, and tested against validation data. While ML models achieved an average accuracy of >96%, the Support Vector Machine Classification model reached 99.58% accuracy in NP-identification. These ML models were then validated via analyses of NPs in water. In each case, the NPs were rapidly and successfully identified, while remaining in their glass bottle. The detection of NPs in water, this new Raman-ML model successfully detected as little as 1E5 particles/L, which surpasses new, published detection limits of only a few months ago.