FullMS2Former: A transformer model for near-complete prediction of peptide tandem mass spectra

FullMS2Former is a transformer-based machine-learning model for near-complete prediction of peptide HCD spectra. Instead of relying on a fixed feature library containing commonly known fragment types, it annotates canonical ions with variable mass offsets, and learned both known and previously unreported fragments directly from spectra. As a result, it predicts the intensities of commonly known fragments (b-ions, y-ions, a-ions, internal fragments and amino-acid immonium ions), as well as fragments that have not been commonly understood or reported in literature. This study uses a subset of the UniSpec dataset (UniSpec-Datasets.7z) originally published by Lapin, Yan and Dong - Lapin, J., Yan, X., & Dong, Q. (2024). UniSpec: Deep learning for predicting the full range of peptide fragment ion series to enhance the proteomics data analysis workflow. Analytical Chemistry, 96(7), 2783-2790.(https://zenodo.org/records/10052268). Detailed file descriptions are in "FileDescription.xlsx"