Loss of micropollutants on syringe filters during sample filtration: Machine learning approach for selecting appropriate filters

Abstract

Prefiltration before chromatographic analysis is critical in the monitoring of environmental micropollutants (MPs). However, in an aqueous matrix, such monitoring often leads to out-of-specification results owing to the loss of MPs on syringe filters. Therefore, this study investigated the loss of seventy MPs on eight different syringe filters by employing Random Forest, a machine learning algorithm. The results indicate that the loss of MPs during filtration is filter specific, with glass microfiber and polytetrafluoroethylene filters being the most effective (<20%) compared with nylon (>90%) and others (regenerated-cellulose, polyethersulfone, polyvinylidene difluoride, cellulose acetate, and polypropylene). The Random Forest classifier showed outstanding performance (accuracy range 0.81-0.95) for determining whether the loss of MPs on filters exceeded 20%. Important factors in this classification were analyzed using the SHapley Additive exPlanation value and Kruskal-Wallis test. The results show that the physicochemical properties (LogKow/LogD, pKa, functional groups, and charges) of MPs are more important than the operational parameters (sample volume, filter pore size, diameter, and flow rate) in determining the loss of most MPs on syringe filters. However, other important factors such as the implications of the roles of pH for nylon and pre-rinsing for PTFE syringe filters should not be ignored. Overall, this study provides a systematic framework for understanding the behavior of various MP classes and their potential losses on syringe filters.

Keywords: Environmental micropollutants; Functional groups; Nylon filter; Prefiltration; Random forest analysis.