[Plastics] A new method for analyzing the chemical nature of microplastics developed using artificial intelligence

© Noëlie Pansiot / Fondation Tara Océan

Until now, the analysis required scientists to manually sort microplastics to identify materials and composition. A time-consuming and tedious method.

Using the 80,000 microplastics collected during the 2014 Tara Mediterranean expedition, a new method based on artificial intelligence was developed. Published in Chemosphere in 2019, this technique automatically determines the chemical nature of microplastics using spectrometry, that is the measurement of light absorbed by plastics. This achievement represents a major advance and constitutes the first part of software, entitled POSEIDON, dedicated to studying microplastics.

Fondation Tara© Noëlie Pansiot / Fondation Tara Océan

Based on artificial intelligence, this method was implemented through tests carried out using a database composed of 969 spectra of microplastics. Results showed that the machine learning process is very effective to identify conventional polymers, such as polyethylene, polypropylene or PET (used in particular to produce water and soda bottles, etc.).

Soon, Tara’s database will be extended to types of less common microplastics and the algorithm coupled to other techniques. In terms of reliability and reproducibility, hopes are high. This method was applied to more than 4,000 types of unidentified microplastics. Verification protocol showed a difference of less than 10% between the results issued by the proposed automated method and human expertise. This discrepancy will soon be reduced since 3/4 can be easily corrected. Automated identification proves to be fast and reliable, even when studying thousands of spectra.

This innovative technique will allow a faster analysis and response to chemical components, toxic to marine biodiversity and human health.

Publication: Mikaël Kedzierskia; Mathilde Falcou-Préfola; Marie-Emmanuelle Kerrosb; Maryvonne Henryc; Maria-Luiza Pedrottib, and Stéphane Bruzauda

a University of Southern Brittany, UMR CNRS 6027, IRDL, F-56100, Lorient, France
b Paris Sorbonne University, UMR CNRS 7093, LOV, F-06230, Villefranche-sur-Mer, France
c IFREMER, LER / PAC, F-83500, Seine-sur-Mer, France

Related articles