High-throughput mass spectrometry has generated repositories containing billions of peptide and small-molecule spectra. Beyond their immediate analytical value, these repositories provide an unprecedented opportunity to uncover fundamental biophysical principles when combined with modern bioinformatics and AI. In this talk, I will illustrate how large-scale reanalysis and synthesis of public data can transform raw spectra into molecular insight. First, I will show how deep contrastive learning embeds hundreds of millions of spectra into a shared latent space, where spectra from the same peptide naturally cluster. Strikingly, this representation captures intrinsic biophysical parameters—including precursor mass, charge state, and sequence composition—without explicit supervision. The learned embedding reveals latent order in fragmentation processes and enables systematic exploration of the “dark proteome.” Second, I will introduce a language-model approach to de novo peptide sequencing that treats mass spectra as a symbolic language of fragment ions. By directly translating spectra into amino acid sequences, the model learns the physics of peptide fragmentation from data alone. In addition to outperforming prior methods, it demonstrates that complex, sequence-dependent fragmentation pathways can be captured through large-scale, data-driven modeling. Finally, I will highlight the construction of a repository-scale “suspect library” that links hundreds of millions of spectra into molecular networks. By propagating structural information across related spectra, this approach uncovers families of metabolites connected by systematic chemical modifications. The resulting network maps expose how structural diversity emerges across metabolism, natural products, and environmental chemistry. Together, these examples illustrate how coupling large-scale public data with powerful AI approaches enables us to extract generalizable physical and chemical principles that govern molecular behavior in biological systems.
Wout Bittremieux (Sun,) studied this question.