Datasets:

Wilhelmlab
/

detectability-proteometools

license: cc-by-4.0
dataset_info:
  features:
    - name: Sequences
      dtype: string
    - name: Classes
      dtype: int64
  splits:
    - name: train
      num_bytes: 6087276
      num_examples: 236758
    - name: val
      num_bytes: 1521645
      num_examples: 59190
    - name: test
      num_bytes: 845949
      num_examples: 32884
  download_size: 5786856
  dataset_size: 8454870
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: val
        path: data/val-*
      - split: test
        path: data/test-*

Detectability - ProteomeTools

This data was collected from the PRIDE repository with the identifiers PXD004732, PXD010595, and PXD021013. The datasets were originally obtained by analyzing pools of approximately 1000 synthetic peptides. RAW data was analyzed using either specific, semi-specific, or unspecific in silico digestion settings in MaxQuant and with Trypsin, LysN, or AspN as specified protease. In all studies, peptide pools were subjected to liquid chromatography using a Dionex 3000 HPLC system (Thermo Fisher Scientific) coupled inline with an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific).

Dataset Details

Curated by: Aalborg University - Denmark, in collaboration with Wilhelmlab - TU Munich.

Dataset Sources

The data is based on the ProteomeTools datasets introduced in [1] and [2] and available at:

Uses

The dataset is intended to be used for training, fine-tuning, and evaluating detectability prediction models, given a peptide sequence. Note that since the peptides were synthesized, training on the dataset will be somewhat biased to synthesizability, rathen than digestability.

References

[1] Zolg, D. P., Wilhelm, M., Schnatbaum, K., Zerweck, J., Knaute, T., Delanghe, B., ... & Kuster, B. (2017). Building ProteomeTools based on a complete synthetic human proteome. Nature methods, 14(3), 259-262.‏

[2] Gessulat, S., Schmidt, T., Zolg, D. P., Samaras, P., Schnatbaum, K., Zerweck, J., ... & Wilhelm, M. (2019). Prosit: proteome-wide prediction of peptide tandem mass spectra by deep learning. Nature methods, 16(6), 509-518.‏

Citation

BibTeX:

@article {Abdul-Khalek2024.10.28.620610,
    author = {Abdul-Khalek, Naim and Picciani, Mario and Wimmer, Reinhard and Overgaard, Michael Toft and Wilhelm, Mathias and Echers, Simon Gregersen},
    title = {To fly, or not to fly, that is the question: A deep learning model for peptide detectability prediction in mass spectrometry},
    elocation-id = {2024.10.28.620610},
    year = {2024},
    doi = {10.1101/2024.10.28.620610},
    publisher = {Cold Spring Harbor Laboratory},
    URL = {https://www.biorxiv.org/content/early/2024/10/31/2024.10.28.620610},
    eprint = {https://www.biorxiv.org/content/early/2024/10/31/2024.10.28.620610.full.pdf},
    journal = {bioRxiv}
}

APA:

Abdul-Khalek, N., Picciani, M., Wimmer, R., Overgaard, M. T., Wilhelm, M., & Gregersen Echers, S. (2024). To fly, or not to fly, that is the question: A deep learning model for peptide detectability prediction in mass spectrometry. bioRxiv, 2024-10.‏

Dataset Card Contact

Simon Gregersen, sgr@bio.aau.dk, Department of Chemistry and Biosciences, Aalborg University.

Mathias Wilhelm, mathias.wilhelm@tum.de, Wilhelmlab, TU Munich, School of Life Sciences, Germany.