DbyDeep

DbyDeep predicts peptide detectability in mass spectrometry (MS)-based proteomics by using a Long Short-Term Memory (LSTM) deep learning model that integrates peptide sequence and protease cleavage-site contexts.

Key Features:

• End-to-End LSTM Network Model: Employs a Long Short-Term Memory (LSTM) network to capture temporal dependencies and sequence contexts for peptide detectability prediction.
• Incorporation of Cleavage Site Contexts: Integrates protease cleavage-site information to account for effects of digestion and sample preparation on peptide detectability.
• Comprehensive Contextual Analysis: Combines sequence context and cleavage-site context to reflect multifactorial influences on detectability in MS experiments.
• Performance Across Diverse MS/MS Data Sets: Demonstrates robust prediction of detectable peptides across MS/MS data sets from multiple species and different mass spectrometry instruments.

Scientific Applications:

• Targeted Proteomics: Informs selection of peptides for targeted proteomics assays by predicting detectability.
• Sample Preparation and Digestion Optimization: Guides optimization of protease digestion and sample preparation through cleavage-site-informed detectability predictions.
• Mass Spectrometry Method Selection: Assists refinement of mass spectrometry settings by identifying peptides likely to be detectable on different instruments.
• MS/MS Data Interpretation: Supports interpretation of heterogeneous MS/MS data sets across species and instrument types by providing detectability likelihoods.

Methodology:

The model is trained end-to-end as an LSTM on extensive MS/MS datasets using peptide sequences and cleavage-site information to learn patterns associated with detectable peptides under various experimental conditions.