Raw signal processing and graph-based visualization to autonomously interpret large repositories of GC-MS data: applications to oil spill weathering studies

Abstract:

Data generated by analytical technology such as gas chromatography (GC), mass spectrometry (MS) and combinations thereof present an interesting dichotomy between large-scale machine interpretation and expert domain knowledge. Machine-interpretation of such data using automated chemometric analysis is typically compound-agnostic and focused on statistical trends (e.g. principal component analysis) of the hundreds, sometimes thousands, of compounds within the sample. On the other hand, expert-annotated compound-cognizant interpretation of the same data is essential to the testing of scientific hypothesis, but by nature is manually driven and therefore, expensive in personnel time, focused only towards a few well-known (target) compounds, and ultimately, limited in its interpretational scope over large data repositories. Automated peak-cognizant processing of the raw instrumental signal bridges the gap between these two complimentary approaches to data interpretation. In particular, peak-cognizant automated interpretation when visualized in a graph setting can provide new opportunities to discover unknown (non-target) compound peaks as well as quantify objectively associations between target and non-target peaks within the raw instrument signal. In this work, we will build upon prior and ongoing work in this computational domain and present new findings from quantifying the environmental weathering of toxic compounds from an oil spill. Specifically, we will present quantitative peak-cognizant visualization of weathering patterns across different toxic contaminants using raw GC-MS signals from a representative sample portfolio collected from the Deepwater Horizon oil spill, Gulf of Mexico, 2010.