Application of a Novel Hybrid CNN-GNN for Peptide Ion Encoding

Area of Science:

• Proteomics
• Bioinformatics
• Computational Biology

Background:

• Machine learning models are crucial for de novo peptide sequencing in mass spectrometry (MS).
• Performance of these models is negatively impacted by inherent MS challenges like spectral noise and missing peaks.
• Current algorithms encode fragment peaks to identify amino acids.

Purpose of the Study:

• To evaluate encoding modules from three state-of-the-art de novo peptide sequencing algorithms.
• To propose and assess a novel convolutional neural network-graph neural network (CNN-GNN) model for encoding peptide ions in MS/MS spectra.
• To compare the proposed model against existing methods in identifying b-ions and y-ions.

Main Methods:

• Extraction and evaluation of encoding modules from existing de novo peptide sequencing algorithms.
• Development of a CNN-GNN model for encoding peptide ions.
• Comparative analysis of encoding module performance on real and artificial MS/MS data.
• Assessment across varying levels of noise and missing peaks.
• Utilized metrics include area under the receiver operating characteristic curve (AUC) and average precision.

Main Results:

• The proposed CNN-GNN model demonstrated superior performance across all tested datasets and metrics.
• Performance was evaluated comprehensively on both real and artificial data with varying noise and missing peak conditions.
• The study highlighted the impact of incorporating additional features, such as intensity rank, into encoding modules.
• Identified limitations associated with using AUC as a sole performance metric were discussed.

Conclusions:

• The developed CNN-GNN model represents a significant advancement for encoding peptide ions in MS/MS spectra.
• This research provides a foundational step towards developing more robust de novo peptide identification algorithms.
• Findings offer valuable insights for future algorithm design in computational proteomics.