Annotating Tandem Mass Spectra in Metabolomics with DreaMS and MassSpecGym.

We present a simple, practical workflow for annotating tandem mass spectrometry (MS/MS) data of small molecules using advanced machine learning and a reference library. The discovery and structural characterization of small molecules is central to biology, chemistry, and medicine. MS/MS has become the leading technology for this task, yet the majority of spectra from untargeted metabolomics remain unannotated, limiting our understanding of the underlying chemical diversity. Artificial intelligence (AI) is rapidly reshaping many areas of science, and mass spectrometry is no exception. In this chapter, we introduce a practical workflow for annotating MS/MS data using two complementary resources: DreaMS, a foundation model for mass spectrometry, and MassSpecGym, a large reference library and benchmarking platform. DreaMS embeds mass spectra into a representation space where structural similarity is preserved, and MassSpecGym provides anchor points for retrieving identical or related molecules. Together, they form an ecosystem that enables efficient annotation of experimental MS/MS data. We demonstrate this approach on alkaloid data from Piper fimbriulatum, showing how DreaMS can retrieve exact matches or place unannotated spectra alongside structurally related compounds. All steps are provided as easy-to-follow protocols, designed to run on standard hardware and requiring no modifications to existing metabolomics pipelines. Readers will learn how to integrate AI tools into their own workflows for compound annotation, molecular networking, and discovery. Overall, this chapter aims to empower biologists and chemists to harness machine learning for metabolomics, transforming unannotated spectra into interpretable chemical knowledge.