π-SPECFormer: A Mass Spectrometry-based Foundation Model for Decoding the Dark Proteome

π-SPECFormer, a mass-spectrometry foundation model pre-trained on 291 million spectra, identifies nine high-confidence mutated peptides in lung adenocarcinoma, including six absent from reference databases.

• Why it matters: A large fraction of acquired spectra in proteomics stays unidentified, masking variation such as sequence mutations, non-canonical peptides, and post-translational modifications; supervised-only deep learning limits learning from unannotated spectra.
• What they did: The model was pre-trained using a two-stage framework combining self-supervised learning with PSMs-guided representation learning, then transferred to de novo sequencing, spectrum clustering, and modification detection.
• The result: In proteomic profiles from 103 individuals with lung adenocarcinoma, it found nine mutated peptides, six not present in reference databases, enabling more unbiased decoding beyond the canonical proteome.

AI summary of the abstract · may contain errors · not medical advice · read the paper.