Pattern Similarity Measures Applied to Mass Spectra

Abstract

Mass spectrometry is a core analytical chemistry technique for elucidating the structure and identity of compounds. Broadly, the technique involves the ionization of an analyte and analysis of the resulting mass spectrum, a representation of ion intensity as a function of mass to charge ratios. In this article, the notion of similarity as it applies to mass spectra is explored. In particular, several modes of approximating distances and similarities in patterns are touched upon: ℓ1 and ℓ2 distances, the Wasserstein metric (earth mover’s distance) and cosine similarity derived measures. Concluding the manuscript is a report on the performance of the similarity measures on a small test set of data, followed by a discussion of mass spectral library searching and prospects for quantifying uncertainty in compound identifications leveraging mass spectral similarity.