The abstract is read at the publisher; the summary is JClub's.
Casanovo-TMT, a deep learning de novo sequencer for TMTpro-labeled peptides, predicts the exact peptidoform for 84.7% of TMTpro-labeled spectra in its test set.
- Why it matters: Existing deep learning de novo models cannot sequence isobaric-tag peptides like TMTpro, limiting multiplexed quantitative proteomics when database search is weak.
- What they did: The authors extended Casanovo’s residue vocabulary with TMTpro tags and phosphorylation, fine-tuned on 21 million spectra, and re-ranked beam candidates using agreement with measured precursor mass.
- The result: Fine-tuning lifted exact peptidoform recovery to 84.7% (test) versus 0.3% without it, and discarding lowest-scoring predictions increased precision to 95% while keeping 81.7%; for phosphopeptides, exact peptidoform prediction reached 69.2% in one enriched dataset and 66–72% in three held-out studies.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.