Mass Spectrometry

Mass spectrometry measures ions by their mass-to-charge ratios; tandem spectra record molecular fragments that help identify compounds, while imaging and MALDI-TOF support spatial and microbial analyses. Machine-learning approaches include transformer-based spectrum prediction and retrieval, fingerprint prediction, and generative models such as diffusion and flow matching for reconstructing molecular structures or peptide sequences. These methods are tested on benchmarks such as MassSpecGym, NPLIB1, and peptide sequencing datasets, as well as in metabolomics, clinical microbiology, and instrument-transfer settings. A central challenge is that strong benchmark scores may not transfer across instruments, labs, or chemical domains, and data leakage, uncertain predictions, and mismatches between intermediate objectives and real retrieval tasks can undermine reliability.