SMSNet's training datasets (WCU-MS-BEST, DeepNovo, ProteomeTools)

Training datasets for SMSNet for comparison with DeepNovo (Figure 2 in https://www.biorxiv.org/content/10.1101/667527v1). In .csv files, each row contain information of a peptide-spectrum match. Files ending with _tgt.csv contain the peptide amino acid sequence in the same ordering as the corresponding .csv files. These _tgt.csv files are provided for direct compatibility with SMSNet's training script (https://github.com/korrawe/SMSNet).<br>For the WCU-MS-BEST dataset, each row of the .csv files contains peptide sequence, charge state, observed m/z, PEP score (from MaxQuant's evidence output file), Andromeda search score (from MaxQuant's evidence output file), and the observed MS/MS peaks.<br>For the DeepNovo and ProteomeTools datasets, each row of the .csv files contains peptide sequence, charge state, observed m/z, and the observed MS/MS peaks.<br>DeepNovo dataset consists of peptides identified through high-resolution MS/MS spectra on Q Exactive instruments as described in the DeepNovo article (Tran <i>et al</i>. PNAS 114:8247-8252, 2017). PRIDE accessions for these experiments are PXD005025, PXD004948, PXD004325, PXD004565,PXD004536, PXD004947, PXD003868, PXD004467, and PXD004424.<br>ProteomeTools dataset consists of synthetic peptides identified through high-resolution MS/MS spectra on Orbitrap Fusion Lumos as described in the ProteomeTools article (Zolg <i>et al</i>. Nat Methods 14:259-262, 2017).<br>All datasets have been split into training, validation, and test sets. Peptides that were identified through multiple MS/MS spectra were grouped so that all identifications appear in the same set. The validation and test sets were further de-duplicated so that they contain only one MS/MS spectrum for each unique peptide.