Compound-Cognizant Feature Compression of Gas Chromatographic Data to Facilitate Environmental Forensics

Abstract

We present complementary compound-cognizant data engineering techniques for feature compression and data indexing across two-dimensional gas chromatographic (GC×GC) datasets with petroleum forensics as the primary application. We propose single-linkage clustering of dominant compounds (targets) along with local interpretation across biomarker peak profiles. Our methods enable high-volume data compression, along with robust querying and forensic distinction between similar sources. We validate our techniques against a diverse dataset of thirty-four crude oil injections collected from nineteen distinct sources across the planet, with emphasis on Macon do well, the source of Deepwater Horizon disaster (Gulf of Mexico, April 2010).