Identification of Microbial Strains via 2D Cross-Correlation of LC-MS Data

Mass spectrometry is commonly used in the identification of species present in microbial samples, but the high similarity in the peptide composition between strains of a single species has made analysis at the subspecies level challenging. Prior research in this area has employed methods such as Principal Component Analysis (PCA), the k-Nearest Neighbors' (kNN) algorithm, and Pearson correlation. Previously, 1D cross-correlation of mass spectra has been shown to be useful in the classification of small molecule compounds as well as in the identification of peptide sequences via the SEQUEST algorithm and its variants. While direct application of cross-correlation to mass spectral data has been shown to aid in the identification of many other types of compounds, this type of analysis has not been demonstrated in the literature for the purpose of LC-MS based identification of microbial strains. A method of identifying microbial strains is presented here that applies the principle of 2D cross-correlation to LC-MS data. For a set of N = 30 yeast isolate samples representing 5 yeast strains (K-97, S-33, T-58, US-05, WB-06), high-resolution LC-MS-Orbitrap data were collected. Reference spectra were then generated for each strain from the combined data of each sample of that strain. Sample strains were then predicted by computing the 2D cross-correlation of each sample against the reference spectra, followed by application of correction factors measuring the asymmetry of the 2D correlation functions.