AlphaPeptDeep

AlphaPeptDeep predicts peptide physicochemical and mass spectrometry properties using deep learning models implemented in a modular PyTorch-based framework.

Key Features:

• Deep Learning Prediction Framework: Implements neural network models in PyTorch to predict peptide properties relevant to proteomics.
• Peptide Property Prediction: Predicts peptide retention time, collisional cross section, and fragment ion intensities for mass spectrometry analysis.
• Generic Post-Translational Modification Representation: Encodes post-translational modifications (PTMs) using chemical composition–based representations.
• Transfer Learning Support: Applies transfer learning to adapt pretrained peptide property models to specific experimental datasets.
• Extensible Model Architecture: Supports extension of the framework to predict additional sequence-based peptide properties, including models for HLA peptide identification in data-independent acquisition (DIA) experiments.

Scientific Applications:

• Mass Spectrometry Proteomics: Predicts peptide properties to improve identification and quantification in LC-MS/MS experiments.
• Immunopeptidomics Analysis: Supports prediction of HLA-associated peptides in data-independent acquisition proteomics datasets.
• Post-Translational Modification Studies: Enables modeling of peptides containing diverse post-translational modifications.

Methodology:

AlphaPeptDeep trains and applies PyTorch-based deep learning models to peptide sequence data, encodes post-translational modifications using chemical composition features, predicts retention time, collisional cross section, and fragment ion intensities, and applies transfer learning to adapt pretrained models to specific proteomic datasets.