A Decade of Computational Mass Spectrometry from Reference Spectra to Deep Learning

Abstract

Computational methods are playing an increasingly important role as a complement to conventional data evaluation methods in analytical chemistry, and particularly mass spectrometry. Computational mass spectrometry (CompMS) is the application of computational methods on mass spectrometry data. Herein, advances in CompMS for small molecule chemistry are discussed in the areas of spectral libraries, spectrum prediction, and tentative structure identification (annotation): Automatic spectrum curation is facilitating the expansion of openly available spectral libraries, a crucial resource both for compound annotation directly and as a resource for machine learning algorithms. Spectrum prediction and molecular fingerprint prediction have emerged as two key approaches to compound annotation. For both, multiple methods based on classical machine learning and deep learning have been developed. Driven by advances in deep learning-based generative chemistry, de novo structure generation from fragment spectra is emerging as a new field of research. This review highlights key publications in these fields, including our approaches RMassBank (automatic spectrum curation) and MSNovelist (de novo structure generation).