Unknown-Unknown Analysis: Strategies for Identifying Compounds Not in Libraries Using Single Quadrupole GC/MS

Traditional strategies to identify unknowns not found in GC/MS search libraries may involve high resolution GC/MS and/or GC/MS-MS to identify the molecular formula of the molecular ion and fragment ions to aid in interpretation. If possible, the isolation and concentration of the unknown for additional spectroscopic analysis (e.g., NMR, IR) can further narrow down possible structures. Finally, the unknown compound is synthesized and compared to the proposed structure, a tedious, lengthy, and expensive process. In many cases, we can utilize various software tools, compound databases, and easily measured complementary metrics, such as retention index (RI), to confidently propose a compound structure in a much more efficient manner using only a traditional single quad GC/MS. The general workflow first requires the calibration of the profile mode spectral data for high mass and spectral accuracy (1) followed by calibrating the GC for retention index. The first step allows for the formula determination of the molecular ion (if it exists) and many of the fragment ions, like high resolution MS. The second step produces an additional confirmational metric to help filter out the many possible structures derived from the molecular formula using AI-predicted RI values (2) calculated from the structures. Both can be done in a simple, highly automated way. Finally, we can use software tools such as the NIST Hybrid Search to match the compound if it is different by a single moiety from spectra in the library and NIST MSInterpreter to further validate the mass spec of the “Unknown-Unknown.” We will describe the process in detail and show real-world examples of its application. (1) The Concept of Spectral Accuracy for MS, Wang and Gu, Anal Chem 2010 82 (17), 7055-7062 (2) AIRI: Predicting Retention Indices and Their Uncertainties Using Artificial Intelligence, Geer, Stein, Mallard, and Slotta, Journal of Chemical Information and Modeling 2024 64 (3), 690-696