DbyDeep: Exploration of MS-Detectable Peptides via Deep Learning

Area of Science:

• Proteomics
• Computational Biology
• Mass Spectrometry

Background:

• Peptide detectability prediction is vital for mass spectrometry (MS)-based proteomics, especially targeted proteomics.
• Current machine learning methods often rely only on peptide sequence or physicochemical properties.
• Peptide detection is influenced by experimental factors beyond sequence, including sample prep, digestion, separation, ionization, and MS parameters.

Purpose of the Study:

• To develop an improved computational model for predicting peptide detectability in mass spectrometry.
• To incorporate contextual information beyond peptide sequences into the prediction model.
• To assess the performance of the new model against existing methods.

Main Methods:

• Developed DbyDeep, an end-to-end Long Short-Term Memory (LSTM) network.
• Incorporated peptide sequence contexts and protease cleavage site contexts into the deep learning model.
• Evaluated DbyDeep on multiple MS/MS datasets across diverse species and MS instruments.

Main Results:

• DbyDeep demonstrated superior performance in predicting peptide detectability compared to existing methods.
• The inclusion of cleavage site contexts significantly improved prediction accuracy.
• The model successfully predicted peptides from diverse datasets, indicating robustness.

Conclusions:

• A comprehensive learning model incorporating multiple contextual factors is necessary for accurate peptide detectability prediction.
• DbyDeep offers a more effective approach by considering sequence and cleavage site contexts.
• The findings highlight the importance of experimental context in MS-based proteomics analysis.