DeepSeqPanII

DeepSeqPanII predicts peptide-HLA class II binding affinities using an end-to-end recurrent neural network with attention to model peptide–HLA class II interactions.

Key Features:

• Recurrent Neural Network with Attention: Employs a recurrent neural network architecture enhanced by an attention mechanism to focus on sequence regions relevant for binding.
• Interpretable Predictions: Uses the attention mechanism to identify and highlight binding cores within peptides and HLA sequences for interpretability.
• Sequence-Based Pan-Specific Model: Provides pan-specific predictions across diverse HLA class II alleles from sequence information without requiring allele-specific training data.
• End-to-End Architecture: Operates as an end-to-end model that does not require explicit pre-processing or post-processing of input sequences.

Scientific Applications:

• Vaccine Development: Predicts peptide binders to HLA class II molecules to support selection of candidate epitopes for vaccine research.
• Autoimmune Disease Research: Facilitates analysis of peptide–HLA interactions relevant to autoimmune disease mechanisms.
• Cancer Immunotherapy: Aids identification of neoantigens that bind HLA class II molecules for cancer immunotherapy design.

Methodology:

Implements an end-to-end recurrent neural network with an attention mechanism and was validated using leave-one-allele-out cross-validation.