Automated non-targeted GC-MS data analysis

Non-targeted analysis has become essential in analytical chemistry, including metabolomics as well as environmental and food analysis. By capturing signals from both known and unknown compounds, non-targeted workflows provide a comprehensive and unbiased view of sample composition.
While LC–HRMS is widely used for non-targeted screening, GC–MS with unit mass resolution remains highly relevant for profiling and fingerprinting due to the excellent separation performance of GC and the broad availability and low cost of quadrupole instruments. However, data processing is still a major bottleneck: baseline correction, feature detection, and retention time alignment are often error-prone and typically require extensive manual intervention.
Here, we present a supervised and an unsupervised automated non-targeted GC–MS data analysis strategy that eliminates the need for feature detection and retention time alignment. The workflows are based on segmentation of chromatograms along the retention time axis, followed by algebraic transformation of matrices and multiway decomposition of transformed segments and/or supervised classification on the decomposed tensor [1, 2].
To enable broad accessibility, we implemented the supervised approach in an interactive browser-based application designed for users without programming experience. The application was developed using the open-source Python packages Bokeh and HoloViews.

Literature:
[1] J. Vestner, G. de Revel, S. Krieger-Weber, D. Rauhut, M. du Toit, A. de Villiers, Toward automated chromatographic fingerprinting: A non-alignment approach to gas
chromatography mass spectrometry data. Acta Chimica Acta 911 (2016) 42-58
[2] K. Sirén, U. Fischer, J. Vestner, Automated supervised learning pipeline for nontargeted GC-MS data analysis. Analytica Chimica Acta: X 1 (2019) 100005