MARIA

MARIA (Major Histocompatibility Complex Analysis with Recurrent Integrated Architecture) predicts the likelihood of antigen presentation by human leukocyte antigen (HLA) class II molecules to support identification of peptides relevant to vaccine development and cancer immunotherapy.

Key Features:

• Multimodal Recurrent Neural Network Architecture: MARIA employs a recurrent neural network that integrates multiple data modalities to predict HLA class II antigen presentation.
• Diverse Training Data Integration: The model is trained on peptide HLA ligand sequences identified by mass spectrometry, antigen gene expression levels, and protease cleavage signatures.
• Superior Predictive Performance: Validation datasets report area under the curve (AUC) values of 0.89–0.92.
• Immunogenic Epitope Identification: MARIA predicts high-scoring peptides that correlate with CD4+ T cell activation and can be used to identify immunogenic epitopes in cancers and autoimmune diseases.

Scientific Applications:

• Vaccine Development: Identification of antigenic peptides presented by HLA class II molecules to inform selection of candidate vaccine epitopes.
• Cancer Immunotherapy: Prediction of neoantigens and peptides likely to be presented by HLA-II to guide personalized immunotherapy strategies and neoantigen prioritization.
• Autoimmune Disease Research: Identification of HLA-II-presented epitopes implicated in autoimmune responses to inform studies of disease-relevant antigens.

Methodology:

A multimodal recurrent neural network was trained on mass spectrometry-derived peptide HLA ligand sequences, antigen gene expression levels, and protease cleavage signatures and evaluated on validation datasets reporting AUC 0.89–0.92.