An open-access computational fingerprinting workflow for source classifications of neat gasoline using GC × GC-TOFMS...

Area of Science:

• Forensic Science
• Analytical Chemistry
• Computational Chemistry

Background:

• Multidimensional chromatography advances gasoline source identification in arson cases.
• Large, complex datasets from gasoline analysis pose data management and interpretation challenges.
• Robust computational techniques are needed for effective gasoline source tracking.

Purpose of the Study:

• Develop a novel, open-access computational fingerprinting workflow for gasoline source tracking.
• Create a regional database of gasoline profiles for arson investigations.
• Enhance the characterization and identification of gasoline sources.

Main Methods:

• Utilized multidimensional gas chromatography-time of flight mass spectrometry (GC × GC-TOFMS) on 69 neat gasoline samples.
• Applied data reduction, normalization, clustering, feature selection (recursive feature addition - RFA), and supervised machine learning (ML).
• Employed decision tree-based ML classifiers with RFA for differentiating gasoline sources.

Main Results:

• Identified 50 key chemical features (n-alkanes, alkenes, cycloalkanes, aromatics) differentiating local gas stations.
• Achieved an average 18% improvement in ML accuracy using RFA with decision tree classifiers compared to using all features.
• Demonstrated the workflow's effectiveness in distinguishing gasoline sources despite clustering overlaps.

Conclusions:

• The open-source computational workflow provides transparency and reproducibility for gasoline source distinction.
• The method aids forensic analysts in identifying ignitable liquids used in wildfire arson.
• The workflow facilitates integration into existing protocols like ASTM E1618-19 without extensive retraining.