Department Environmental Chemistry

RÜS Data 3.0 - AI-guided characterization of non-target substances

Photo: Rosi Siber, Eawag
Map of the Rhine River sampling station at Basel with the river catchment


The river Rhine is a crucial water resource for Europe, providing drinking water to millions of people. However, the Rhine is one of the most human-impacted rivers with high industrial activity on the riverbank, which requires close monitoring. To that aim, RÜS (Rheinüberwachungsstation Weil am Rhein) measures daily intensities of hundreds of known and thousands of unknown substances via LC-HRMS/MS. The combined use of target and non-target screening techniques helps to shift the focus from a few legacy chemicals to complex mixtures of unknown substances.

In a previous project (Rues Data 2.0), we showed that 50% of the prioritized substances detected in the Rhine near Basel may originate from industry. We now aim for a deeper understanding of these important contaminants.

The aim of the RUES Data 3.0 project is therefore to perform a multifaceted characterization of pollutants to better comprehend their behavior and potential environmental effects. This workflow is organized into three work packages:

WP1 – Extended characterization of industrial emissions: A detailed descriptor for industrial emissions is developed that uses numerical metrics to characterize the time profile dynamics of the contaminants. This descriptor allows simple prioritization of industrial emission profiles based on the duration, intensity, and course of their occurrence.

WP2 – Data-driven Chemical Characterization: Candidates for chemical characterization of unknown contaminants are identified to enhance our understanding of the type of pollutants present in the chemical space. This is achieved via automatic processing pipelines basing on the information recorded from the MS1 and MS2 spectra.

WP3 – AI-Powered Toxicity Estimation: an in-silico toxicity estimate is calculated using machine learning to evaluate the potential toxicity of unknown compounds.  This approach is implemented through the framework MLinvitroTox.

The results of the three work packages lead to a comprehensive chemical, toxicological and source-based evaluation of the measured contaminants in the Rhine.

Publications

Unveiling industrial emissions in a large European river: insights from data mining of high-frequency measurements

Despite the tremendous efforts to improve river water quality, chemical contamination remains a significant issue. Besides well-known contaminants, in recent years, pollutants of industrial origin received increasing attention because of the huge knowledge gap regarding their occurrence, fate and environmental risks. Moreover, such pollutants often exhibit high concentration fluctuations over time, which makes them less predictable and measurable with classical short-time campaigns.
This study provides insights into the different sources of chemical contamination of the Rhine River based on temporal high-frequency LC-HRMS monitoring data from a single location. A newly developed prioritization strategy selected nearly 3000 substances as potentially major contaminants. A novel classification analysis based on temporal behavior identified 53 % of these compounds (accounting for 62 % of the time-integrated intensity recorded in the dataset) as originating from irregular emission sources. Irregular emissions can originate from industrial production cycles. After delimiting other potential irregular sources, we have strong evidence indicating that a considerable share of the irregular emissions likely comes from industrial activities. This finding is supported by the structural elucidation of sixteen irregularly emitted substances, for which the industrial origin was successfully confirmed. Those compounds include 3-chloro-5-(trifluoromethyl)pyridine-2-carboxylic acid and 4-(dimethylamino)-2,2-diphenylpentanenitrile. In addition, 40 other compounds exhibited temporal emission patterns similar to the sixteen industrial compounds, which strongly suggests a common contamination source. Finally, 100 top-ranking compounds were selected for further structural elucidation and emission reduction measures. The computational approach outlined within this study can be effectively applied in other large river catchments to identify unknown contaminants stemming from industrial sources.
Chonova, T.; Ruppe, S.; Langlois, I.; Griesshaber, D. S.; Loos, M.; Honti, M.; Fenner, K.; Singer, H. (2025) Unveiling industrial emissions in a large European river: insights from data mining of high-frequency measurements, Water Research, 268, 122745 (13 pp.), doi:10.1016/j.watres.2024.122745, Institutional Repository