Employing and Interpreting a Machine Learning Target-Cognizant Technique for Analysis of Unknown Signals in Multiple...

Area of Science:

• Analytical Chemistry
• Environmental Science
• Computational Science

Background:

• Gas chromatography-mass spectrometry (GC/MS/MS) generates complex raw data with numerous signal peaks.
• Accurate identification and association of chemical compounds, both expected (target) and unexpected (non-target), are crucial for sample analysis.
• Polychlorinated biphenyls (PCBs) are environmental contaminants requiring robust detection and spatial analysis.

Purpose of the Study:

• To develop a robust signal processing and machine learning framework for autonomous peak association in GC/MS/MS data.
• To evaluate the accuracy of different machine learning algorithms in classifying and associating signal peaks.
• To visualize the geographical distribution and trends of target PCBs in air samples.

Main Methods:

• Adaptive signal processing for data smoothing, baseline correction, and separation of coeluted peaks.
• Application of three machine learning algorithms for autonomous peak association, with a focus on a random forest classifier.
• Analysis of 150 air samples using GC/MS/MS to detect individual PCBs across diverse Chicago locations.
• Utilizing 80% of data for training and 20% for testing the machine learning models.

Main Results:

• The study successfully developed and tested a machine learning framework for autonomous peak association in GC/MS/MS data.
• The random forest classifier demonstrated effectiveness in associating identified peaks with target PCB peaks.
• Topographical illustrations revealed distinct patterns of PCB accumulation and degradation across different Chicago locations.
• The framework enabled the association of non-target peaks with known target PCBs, enhancing sample analysis.

Conclusions:

• The developed interdisciplinary framework provides a robust method for signal processing and autonomous machine learning in GC/MS/MS analysis.
• This approach improves the identification and association of both target and non-target chemical compounds.
• The geographical mapping of PCBs offers valuable insights into environmental contamination patterns and dynamics.