Annotation of MS/MS spectra remains highly unsolved

Establish computational methods that achieve reliable, high-accuracy annotation of tandem mass spectra (MS/MS) with correct molecular structures at scale, closing the performance gaps observed for baseline models in the MassSpecGym benchmark.

Background

The MassSpecGym benchmark evaluates de novo structure generation, molecule retrieval, and spectrum simulation under a generalization-demanding split. Baseline models show limited performance, particularly for de novo generation where accuracy is zero, indicating substantial room for improvement.

Based on these results, the authors explicitly state that annotating MS/MS spectra with molecular structures remains a highly unsolved problem, motivating further methodological advances and standardized benchmarking.