---
license: cc-by-4.0
dataset_info:
- config_name: default
  features:
  - name: raw_file
    dtype: string
  - name: scan_number
    dtype: int64
  - name: method_nbr
    dtype: int64
  - name: precursor_charge_onehot
    sequence: int32
  - name: collision_energy_aligned_normed
    dtype: float64
  - name: intensities_raw
    sequence: float64
  - name: package
    dtype: string
  - name: modified_sequence
    dtype: string
  splits:
  - name: train
    num_bytes: 32998969225
    num_examples: 21294649
  - name: val
    num_bytes: 9419944436
    num_examples: 6078851
  - name: test
    num_bytes: 4654383726
    num_examples: 3003623
  download_size: 3706447347
  dataset_size: 47073297387
- config_name: holdout
  features:
  - name: raw_file
    dtype: string
  - name: scan_number
    dtype: int64
  - name: modified_sequence
    dtype: string
  - name: collision_energy_aligned_normed
    dtype: float64
  - name: intensities_raw
    sequence: float64
  - name: precursor_charge_onehot
    sequence: int32
  - name: method_nbr
    dtype: int64
  splits:
  - name: test
    num_bytes: 1208775037
    num_examples: 783150
  download_size: 50335479
  dataset_size: 1208775037
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: val
    path: data/val-*
  - split: test
    path: data/test-*
- config_name: holdout
  data_files:
  - split: test
    path: holdout/test-*
---

# PROSPECT PTMs - Fragment Ion Intensity Prediction (MS2)

A mass-spectrometry dataset for applied machine learning in proteomics, annotated, processed and split for the task of fragment ion intensity prediction.

## Dataset Details

- **Curated by:** Wilhelmlab - Technical University of Munich - School of Life Sciences - Germany
- **License:** CC-BY4.0

### Dataset Sources

The data is based on the PROSPECT PTMs datasets hosted in Zenodo [3][4][5][6]. 
- **Repository:** https://github.com/wilhelm-lab/PROSPECT

## Uses

The dataset is intended to be used for fragment ion intensity prediction given a peptide sequence. Optionally, addiional inputs can be used as features to encode PTMs or other features describing the experimental setup.

## Dataset Structure

The dataset has two configurations:
- default: this is the train/val/test data that is based on all the Zenodo PROSPECT PTMs datasets [3][4][5]
- holdout: this is the holdout datasets that can be eventually used to evaluate a model capable of processing PTMs. It is based on the PROSPECT Test-PTM dataset on Zenodo [6].

Use one of the following lines to load the respective configuration:

```python
# main data for training and evaluation; contains train, val, tesst splits
main_dataset = load_dataset("Wilhelmlab/prospect-ptms-ms2")

# holdout data for final benchmarking and model comparison; contains test split only
holdout_dataset = load_dataset("Wilhelmlab/prospect-ptms-ms2", "holdout")
```


## Dataset Creation

### Curation Rationale

The dataset is intended to serve as a reference benchmark dataset for fragment ion intensity prediction, processed, split, and ready-to-use for developing deep learning models for this specific task.


### Source Data

The upstream source data is based on the ProteomeTools datasets available on PRIDE [1][2].

#### Data Collection and Processing

[More Information Needed]


### Annotations

The annotations are based on an expert system [7] with a set of rules listed in the PROSPECT paper [8]. The vector of intensities is collected in one column named `intensities_raw`.


#### Personal and Sensitive Information

The dataset does not contain any personal, sensitive, or private data.

### Recommendations

We recommend using the holdout configuration for solely evaluation models at the end of the research iteration. 

## Citation

**BibTeX:**

[More Information Needed]

**APA:**


## References

[1] Daniel P Zolg, Mathias Wilhelm, Karsten Schnatbaum, Johannes Zerweck, Tobias Knaute, Bernard Delanghe, Derek J Bailey, Siegfried Gessulat, Hans-Christian Ehrlich, Maximilian Weininger, et al. Building proteometools based on a complete synthetic human proteome. Nature methods, 14(3):259–262, 2017.

[2] Daniel Paul Zolg, Mathias Wilhelm, Tobias Schmidt, Guillaume Médard, Johannes Zerweck, Tobias Knaute, Holger Wenschuh, Ulf Reimer, Karsten Schnatbaum, and Bernhard Kuster. Pro- teometools: Systematic characterization of 21 post-translational protein modifications by liquid chromatography tandem mass spectrometry (lc-ms/ms) using synthetic peptides. Molecular & Cellular Proteomics, 17(9):1850–1863, 2018.

[3] Wassim Gabriel, Omar Shouman, and Mathias Wilhelm. PROSPECT PTMs Dataset - Multi- PTM. DOI:https://doi.org/10.5281/zenodo.11472525, 2024.

[4] Wassim Gabriel, Omar Shouman, and Mathias Wilhelm. PROSPECT PTMs Dataset - TMT. DOI:https://doi.org/10.5281/zenodo.8221499, 2023.

[5] Wassim Gabriel, Omar Shouman, and Mathias Wilhelm. PROSPECT PTMs Dataset - TMT- PTM. DOI:https://doi.org/10.5281/zenodo.11474099, 2024.

[6] Wassim Gabriel, Omar Shouman, and Mathias Wilhelm. PROSPECT PTMs Dataset - Test-PTM. DOI:https://doi.org/10.5281/zenodo.11477731, 2024.

[7] Nadin Neuhauser, Annette Michalski, Jürgen Cox, and Matthias Mann. Expert system for computer-assisted annotation of ms/ms spectra. Molecular & Cellular Proteomics, 11(11):1500– 1509, 2012.

[8] Omar Shouman, Wassim Gabriel, Victor-George Giurcoiu, Vitor Sternlicht, and Mathias Wil- helm. PROSPECT: Labeled tandem mass spectrometry dataset for machine learning in pro- teomics. In S. Koyejo, S. Mohamed, A. Agarwal, D. Belgrave, K. Cho, and A. Oh, editors, Advances in Neural Information Processing Systems, volume 35, pages 32882–32896. Curran Associates, Inc., 2022.


## Dataset Card Contact

mathias.wilhelm@tum.de

Wilhelmlab, TU Munich, School of Life Sciences, Germany.