Deep learning-driven fragment ion series classification enables highly precise and sensitive de novo peptide sequencing

This Zenodo record contains the dataset and model weights for "Deep learning-driven fragment ion series classification enables highly precise and sensitive de novo peptide sequencing".

This repository contains the following files:

• For the human dataset by Wang et al.:

  • train_val_test_split.csv containing the mapping of the correct peptide by MaxQuant to either train, validation or test set

  • psms_train_val_test.csv containing the mapping of correct PSMs (scan number, raw file and correct peptide by MaxQuant) to either train, validation or test set

  • updated_spectralis_test_out.csv as before containing Spectralis-EA predictions and scores on test set, as well as initial peptides and scores by Casanovo and Novor and now containing also correct peptides by MaxQuant and Spectralis-scores on the combination of Casanovo and Novor sequences (column named spectralis_score_onlyRescoring)

  • spectralis_test_out_heart_analysis.csv subset of 20220822_spectralis_test_out.csv containing only PSMs for the tissue heart with the computation of precision and recall values

  • spectralis_test_out_pointnovo_deepnovo.csv containing predictions by DeepNovo and PointNovo with original scores and Spectralis-score, as well as correct peptides by MaxQuant

• For the nine-species dataset by Tran et al.:

  • spectralis_ninespecies_out.csv containing spectrum identifiers, correct peptides by PEAKSDB, predicted peptides by the different de novo sequencing tools as well as original scores and Spectralis-scores for the different PSMs.