Validation of the Fast Expectation-Maximization Algorithm for Source Apportionment of Organic Compounds Using Nontarget HRMS Fingerprints

Source apportionment of organic pollutants in complex
environmental
systems remains a major analytical challenge, particularly when dealing
with unknown or numerous pollution sources. While high-resolution
mass spectrometry (HRMS) offers rich chemical fingerprint data, few
computational methods can efficiently and accurately resolve multicomponent
sources without prior compositional knowledge. To address this gap,
we introduced the fast expectation-maximization (FEM) algorithm that
enables rapid, reliable, and unsupervised source attribution using
nontarget HRMS data. The FEM algorithm was evaluated under both controlled
laboratory conditions, simulating mixtures of municipal wastewater,
hospital effluent, and urban runoff, and real case. It demonstrated
robust performance, producing accurate contribution estimates (within
0.58–2.61 times theoretical values) even in multisource systems.
The method exhibited accuracy comparable to established dilution curve
methods and significantly outperformed the widely used SourceTracker
tool in both speed and quantitative precision. This work proved the
FEM algorithm as a powerful new computational tool for rapid and high-resolution
pollution source tracking using the large data set of nontarget HRMS
analysis.