Standard approaches to mass-spectral compound identification use algorithms to measure the spectral similarity or dissimilarity between the mass spectra of an unknown compound and knowns in a database. Current algorithms are limited by the variance between replicate mass spectra, and the invalid assumption that the variance in ion abundance at each <i>m/z</i> within a spectrum is independently variable. To investigate this, the relative contributions of different instrument parameters to the observed spectral variance in replicate tandem mass spectra were assessed using cocaine as a test compound. Three sets of replicate spectra were collected for a five-factor, three-level full factorial experiment on an Agilent 6538 UHD QTOF. Results showed the collision energy and fragmentor voltage were responsible for the most variation in the recorded abundances of the ions. The skimmer voltage and capillary voltage were both responsible for a moderate amount of variance, and drying gas temperature was responsible for the least. Combined, these sources explained 99% of the total variance in the replicate spectra. This knowledge then guided further experimentation to attempt to create a binary classifier capable of differentiating between THC and CBD, which are traditionally difficult to distinguish based on their tandem mass spectra. Sixty replicate spectra were collected for each cannabinoid at three different collision energies. Thirty spectra from each cannabinoid were used as a training set on which to develop linear models, and the remaining spectra were used as test spectra. The top 20 most abundant ions in each of the two training sets were used to develop 20 multivariate linear models, where each fragment is iteratively considered to be a dependent variable and the remaining 19 fragments are considered independent variables. Whereas conventional approaches to spectral comparisons could only discriminate the two conformers with 60-80% confidence, depending on the collision energy, the linear modeling approach typically provided over 95% confidence.

Using an algorithm to resolve the tandem mass spectra of THC and CBD

ACS Spring 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.