Spectralis: Deep Learning for De Novo Peptide Sequencing from Mass Spectrometry

Zenodo records contain datasets and model weights for a deep learning method classifying fragment ion series. The data includes human and nine-species mass spectrometry datasets, with files mapping peptides and PSMs for training and evaluation. Daniela Klaproth-Andrade from the Technical University of Munich authored this resource.

Use Cases

• Benchmarking de novo sequencing tools like Casanovo, Novor, DeepNovo, and PointNovo based on provided prediction and score files.
• Training and validating deep learning models for fragment ion series classification using the provided train/validation/test splits.
• Analyzing precision and recall for peptide sequencing in specific tissues, such as heart, using the provided subset file.
• Comparing peptide predictions across different species using the nine-species dataset.

Strengths

• Includes data from two distinct studies: a human dataset by Wang et al. and a nine-species dataset by Tran et al.
• Provides structured CSV files for model training, validation, and testing, including mappings for correct peptides and PSMs.
• Contains predictions and scores from multiple established de novo sequencing tools (Casanovo, Novor, DeepNovo, PointNovo) for comparison.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Zenodo, associated with a research paper from the Technical University of Munich.

Collection Method

Likely derived from mass spectrometry experiments and processed with tools like MaxQuant and PEAKSDB.