Machine Learning for Absolute Quantification of Unidentified Compounds in Non-Targeted LC/HRMS

Area of Science:

• Environmental Chemistry
• Analytical Chemistry
• Mass Spectrometry

Background:

• Liquid chromatography/electrospray ionization/high-resolution mass spectrometry (LC/ESI/HRMS) is vital for water pollutant monitoring, with non-targeted analysis gaining traction.
• Current non-targeted analysis is largely qualitative due to the absence of analytical standards for many detected compounds.
• Existing models for quantitative analysis often require tentatively known structures, which are unavailable for most detected pollutants.

Purpose of the Study:

• To develop a novel quantification approach for LC/ESI/HRMS non-targeted analysis that does not rely on known chemical structures.
• To enable accurate quantification of chemical pollutants even when only mass and retention time data are available.
• To improve the reliability and scope of water quality monitoring through advanced analytical techniques.

Main Methods:

• A new quantification approach was developed utilizing LC/ESI/HRMS descriptors for compounds with unknown structures.
• The method was trained and validated using 92 diverse compounds analyzed across various mobile phase pH conditions and ionization modes (positive and negative ESI).
• Performance was benchmarked against baseline methods: assuming equal response factors and using the closest eluting standard's response factor.

Main Results:

• The developed machine learning-based approach achieved a mean prediction error of only a factor of 10.
• This represents a substantial improvement over the baseline methods, which yielded errors of factors 29 and 1300, respectively.
• Validation on blind samples with 48 spiked compounds showed a mean prediction error below a factor of 6.0, comparable to structure-dependent methods.

Conclusions:

• The developed LC/ESI/HRMS quantification approach effectively quantifies pollutants without requiring prior structural identification.
• This method significantly enhances the accuracy of non-targeted water pollutant analysis, overcoming limitations of current qualitative approaches.
• The findings offer a powerful tool for environmental monitoring, enabling more precise risk assessment and regulatory compliance.