Weakly supervised peptide-TCR binding prediction facilitates neoantigen identification.

Area of Science:

• Computational biology
• Immunology
• Bioinformatics

Background:

• Neoantigen identification is crucial for tumor immunotherapy but computationally challenging.
• Current methods often overlook patient-specific T cell receptor (TCR) profiles, limiting neoantigen immunogenicity evaluation.
• Existing tools face performance and application constraints in real-world neoantigen identification.

Purpose of the Study:

• To develop a novel framework, TCRBagger, for enhanced personalized neoantigen identification.
• To integrate self-supervised, denoising, and multi-instance learning (MIL) for modeling peptide-TCR binding.
• To improve the prediction of immunogenic neoantigens by considering patient-specific TCR profiles.

Main Methods:

• Developed TCRBagger, a weakly supervised learning framework utilizing bagging of sample-specific TCR profiles.
• Integrated three learning strategies: self-supervised learning, denoising, and multi-instance learning (MIL).
• Modeled peptide-TCR binding interactions to identify immunogenic neoantigens.

Main Results:

• TCRBagger demonstrated superior performance compared to existing neoantigen prediction tools.
• The framework effectively models peptide-TCR profile interactions for enhanced immunogenicity identification.
• Improved capability in identifying immunogenic neoantigens for personalized tumor immunotherapy.

Conclusions:

• TCRBagger offers a novel perspective and methodology for modeling peptide-TCR interactions.
• The framework facilitates personalized neoantigen identification for tumor immunotherapy.
• TCRBagger enhances the accuracy and applicability of neoantigen prediction tools.