Towards reliable data: Validation of a machine learning-based approach for microplastics analysis in marine organisms using Nile red staining.

Microplastic (MP) research faces challenges due to costly, time-consuming, and error-prone analysis techniques. Additionally, the variability in data quality across studies limits their comparability. This study addresses the critical need for reliable and cost-effective MP analysis methods through validation of a semi-automated workflow, where environmentally relevant MP were spiked into and recovered from marine fish gastrointestinal tracts (GITs) and blue mussel tissue, using Nile red staining and machine learning automated analysis of different polymers. Parameters validated include trueness, precision, uncertainty, limit of quantification, specificity, sensitivity, selectivity, and method robustness. For fish GITs a 95 ± 9 % recovery rate was achieved, and 87 ± 11 % for mussels. Polymer identification accuracies were 76 ± 8 % for fish GITs and 80 ± 13 % for mussels. Polyethylene terephthalate fragments showed more variability with lower accuracies. The proposed validation parameters offer a step towards quality management guidelines, as such aiding future researchers and fostering cross-study comparability.