Expert Algorithm for Substance Identification Using Mass Spectrometry: Statistical Foundations in Unimolecular...

Area of Science:

• Analytical Chemistry
• Mass Spectrometry
• Chemometrics

Background:

• Accurate identification of organic substances from mass spectra is challenging without contemporaneous analysis on the same instrument.
• Existing methods struggle with variations in spectra due to changing analytical conditions within or between instruments.

Purpose of the Study:

• To develop a robust method for accurate organic substance identification from mass spectra, independent of contemporaneous analysis.
• To establish a reliable approach for identifying cocaine and differentiating it from its diastereomers using mass spectrometry data.

Main Methods:

• Utilized Rice-Ramsperger-Kassel-Marcus (RRKM) theory to predict linear correlations in mass spectral branching ratios under varying conditions.
• Applied general linear modeling (GLM) to a database of 128 cocaine mass spectra, focusing on the 20 most abundant fragments.
• Validated statistical models using analysis of variance (ANOVA) and residual distribution analysis.

Main Results:

• GLM models explained over 90% of the variance in normalized ion abundances for cocaine spectra.
• Models predicted ion abundances with <2% relative error, despite >30% variation in measured abundances, when applied to spectra from over 20 laboratories.
• Residual errors were significantly larger for known negative spectra (including cocaine diastereomers) compared to known positive cocaine spectra.

Conclusions:

• General linear modeling provides a statistically valid and accurate method for identifying organic substances from mass spectra, even across different instruments and laboratories.
• The developed GLM approach enables reliable binary classification and identification of cocaine, effectively distinguishing it from its diastereomers and other known negatives.