BOOK OF ABSTRACTS 185 IHSS&IWA26 / BRNO / CZECHIA / 23–28 August 2026 Hall A Poster Session / Analysis and Characterization P1.5 Deciphering Humic Substance Continuity: Source-Aware Molecular Signatures and Reactomic Structure Revealed by FT-ICRMS and Interpretable Machine Learning Huiyun Xue1*, Kanako Toda1, Takumi Saito1, 1 Department of Nuclear Engineering and Management, School of Engineering, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8656, JAPAN * xuehuiyun@g.ecc.u-tokyo.ac.jp Humic acids (HA) and fulvic acids (FA) are operationally defined fractions of humic substances separated by pH-dependent precipitation, yet the molecular validity of this fractionbased framework remains unresolved. A long-standing question is whether HA and FA should be interpreted as chemically discrete classes or as different organizational states within a shared humic continuum [1]. This question remains important in environmental chemistry because HA/FA terminology is still widely used to discuss mobility, sorption behavior, redox activity, and carbon persistence in natural systems [2]. In this study, we develop a sourceaware analytical framework to reassess the conventional HA/FA classification by integrating ultrahigh-resolution FTICR-MS, molecular formula assignment, spectral alignment, post-assignment molecular interpretation, interpretable machine learning, and reactomics [3-5]. Formula assignment and molecular comparison are grounded in established ultrahigh-resolution DOM workflows [3], while cross-sample spectral harmonization is strengthened through Gaussian-based alignment [4]. Postassignment chemical characterization is supported by Python-based FTMS analysis tools [5], and model interpretation is guided by SHAP to ensure that discriminative patterns can be translated into chemically meaningful molecular signatures rather than treated as black-box outputs. The framework is explicitly designed to address three linked objectives. First, it tests whether HA and FA can be robustly differentiated at the sample level when molecular composition, weighted chemical descriptors, and structural heterogeneity are considered together. Second, it examines whether broad environmental-domain effects and finer source-subtype structure introduce major axes of variation beyond the conventional fraction effect. Third, it evaluates whether formula-level signatures, paired-massdifference connectivity, hub molecules, and transformation families support a model of discrete molecular classes or a model of continuity built on shared
RkJQdWJsaXNoZXIy NDA4Mjc=