IHSS&IWA26 / BRNO / CZECHIA / 23–28 August 2026 BOOK OF ABSTRACTS 32 Analysis and Characterization Monday, 24 August 2026 / Hall B+C SL10 Mitigation of Batch Effects in High Resolution Mass Spectrometry of Natural Organic Matter Helen Rutlidge1, Amirali Sadeghloo1, Russell Pickford3, Fitri Widhiastuti1, and Rita Henderson1 1 UNSW Sydney, School of Chemical Engineering, High St, UNSW Sydney, NSW 2052, Australia 2 UNSW Sydney, Bioanalytical Mass Spectrometry Facility There is increasing interest in exploring more detailed molecular characterization of Natural Organic Matter (NOM) in water, as changes in NOM composition can affect water treatment performance and the formation of harmful disinfection by-products. High-resolution mass spectrometry is commonly being used for this purpose, with Orbitrap mass spectrometry becoming a more accessible alternative to Fourier transform ion cyclotron resonance mass spectrometry for NOM analysis. However, the high sensitivity of Orbitrap mass spectrometry means that equipment drift and technical variation between analytical runs can strongly influence the resulting data and make samples analyzed on different days difficult to compare directly. The aim of this study was to investigate the extent of this batch-related variation in nontargeted NOM analysis and evaluate strategies to reduce it while preserving meaningful chemical differences. To investigate these effects, a multirun environmental NOM dataset was generated using a Thermo Fisher Q Exactive HF Orbitrap mass spectrometer. The dataset included five environmental water samples, three standards, and three pooled quality-control (QC) sample types distributed across five analytical runs arranged in two blocks separated by two weeks. Data was processed from Compound Discoverer (V3.4) outputs and analyzed using a range of correction strategies, including simple data-driven normalization methods, batch-correction approaches, and QC-based machinelearning methods. Comparisons were made using principal component analysis (PCA), QC relative standard deviation, and feature-level tests to assess both batch effects reduction and preservation of real chemical differences. The results showed that different correction strategies had different strengths. Methods focused mainly on QC correction were effective at improving QC compactness, while broader technical-variation correction methods performed better when both batch reduction and preservation of meaningful sample structure were considered together. Across the tested workflows, Technical variation Elimination with ensemble learninG architecture (TIGER) best reduced batch effects while
RkJQdWJsaXNoZXIy NDA4Mjc=