Complex mixtures — that is, mixtures of individual compounds that may be difficult to differentiate using chromatographic methods — are ubiquitous in many areas of life sciences and chemistry. High-resolution mass spectrometry (MS) is one of the most versatile analysis methods, typically capable of detecting compounds at the trace level and elucidating chemical structures. Even so, characterizing complex mixtures can be remarkably challenging, as they generate complicated spectra that are difficult to interpret. One source of complexity is the presence of isotopologues, molecules that differ only in isotopic composition, which results in groups of many MS signals rather than single signals for individual molecules. Recent work by Valentine Ananikov and colleagues focused on addressing this problem with MEDUSA, a framework for MS data analysis that improves the efficiency of analysis and lessens the problem of isotopologues.
MEDUSA follows a three-step workflow: spectra preprocessing, the analysis itself and visualization. During preprocessing, MEDUSA implements gradient-boosted decision modeling to perform deisotoping, which reduces the complexity of the spectrum by removing unwanted isotopologue peaks. The outcome of the deisotoping algorithm is then fed into neural network classifiers and regressors in the analysis step, in which the analysis of the elements present in the mixture enables searches for a group of compounds of interest. The neural network regressor’s output and the confidence intervals are used to define the search range, and the classifier predictions are used to obtain a list of elements to search, which together substantially reduces the search space and improves the speed of data analysis. Finally, MEDUSA generates PCA, t-SNE and cluster maps for visualization, in order to confirm the presence of a compound in the spectrum. Overall, MEDUSA is able to make three levels of mass spectra analysis (entire spectrum, individual compound and every single compound) possible, making it a highly useful tool for complex mixture characterization. The authors tested MEDUSA on three complex examples (fragment ion annotation of small peptides, analysis of tea extracts, and a study of cross-coupling catalysts in the Sonogashira reaction), demonstrating that the tool can generalize to a wide variety of applications. This fully automated MS data analysis tool improves the efficiency and, therefore, the practicality of analyzing complex mixtures, addressing the long-standing problem of formula determination in untargeted analysis.
This is a preview of subscription content, access via your institution