MassNet: open 1.5-billion-spectrum proteomics corpus and the XuanjiNovo de novo peptide-sequencing model (Nature Methods)

Nature Methods (open access, 14 October 2026) presents MassNet, an AI-ready data corpus of 1.5 billion tandem mass spectra from data-dependent acquisition proteomics across 35 species, a compact MSDT data format, and XuanjiNovo, a non-autoregressive transformer for de novo peptide sequencing (reported 38.35 to 143.57% better than prior methods on human Astral data). The dataset is public in the PRIDE archive (PXD071590, CC0; released 20 October 2026) and the code and weights are on GitHub under Apache-2.0. Proteomics rather than phage; performance figures are the authors'.