MassNet: billion-scale AI-ready mass spectral corpus empowering scalable deep learning in proteomics.

MassNet is a billion-scale AI-ready mass spectral corpus containing 1.5 billion spectra across 35 species, enabling scalable deep learning in proteomics.

• Why it matters: This large, standardized dataset addresses the need for comprehensive, high-quality proteomics data to advance AI applications and improve peptide sequencing accuracy.
• What they did: The authors compiled MassNet with an AI-compatible data format called MSDT and developed XuanjiNovo, a transformer model that uses curriculum learning to improve de novo peptide sequencing, achieving 38.35-143.57% better performance than existing methods on human Astral data.
• The result: MassNet provides a foundational resource for AI-driven proteomics tasks such as retention time prediction and PSM rescoring, facilitating further advancements in the field.