Self-iterative multiple instance learning enables the prediction of CD4+ T cell immunogenic epitopes 1. ImmuScope is an innovative deep-learning framework designed to predict CD4+ T cell immunogenic epitopes with high accuracy. It integrates multi-allelic and single-allelic data to enhance predictions in immune response initiation, making it a key tool for advancing vaccine and immunotherapy research. 2. The model leverages a self-iterative multiple-instance learning (MIL) approach combined with positive-anchor triplet loss, enabling it to handle weakly labeled multi-allelic data efficiently. This approach significantly increases the coverage of alleles while minimizing label ambiguity. 3. ImmuScope outperforms existing methods in multiple tasks, including antigen presentation prediction, MHC-II binding specificity discovery, CD4+ T cell epitope prediction, and immunogenicity assessment. Its robustness and generalizability were validated across diverse datasets, showing superior predictive power. 4. A key breakthrough of ImmuScope is its application in discovering melanoma neoantigens, where it revealed mutations in epitope sequences that affect pMHC-II binding and immunogenicity, providing deeper insights into cancer immunology. 5. The framework was also applied to SARS-CoV-2 epitope analysis, successfully predicting immune escape mechanisms in the Omicron variant and aligning with experimental data on immune response dynamics. 6. ImmuScope's ability to perform motif deconvolution on multi-allelic data provides critical insights into MHC-II binding specificities, aiding in the identification of new epitopes for immunotherapy and vaccine development. 7. This comprehensive solution not only enhances our understanding of T cell activation but also facilitates personalized medicine by identifying immunogenic neoepitopes and optimizing therapeutic interventions. @supercs08 @RossjohnLab @Zachary_Zhikang @yumengzhang99 📜Paper: biorxiv.org/content/10.1101/… #ImmuScope #TCellImmunity #Immunotherapy #VaccineDevelopment #MachineLearning #Immunopeptidomics #CancerImmunology #SARSCoV2 #Neoantigen