MIX-TPI: a flexible prediction framework for TCR-pMHC interactions based on multimodal representations

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• T-cell receptor (TCR) and peptide-major histocompatibility complex (pMHC) interactions are crucial for adaptive immunity.
• Predicting these interactions is challenging due to limited data, data heterogeneity, and high experimental costs.

Purpose of the Study:

• To develop a novel computational framework, MIX-TPI, for predicting TCR-pMHC interactions.
• To leverage amino acid sequences and physicochemical properties for improved prediction accuracy.

Main Methods:

• Utilized convolutional neural networks (CNNs) as the core architecture.
• Incorporated sequence-based and physicochemical-based extractors.
• Employed modality-invariant and modality-specific representations.
• Integrated a self-attention fusion layer for classification.

Main Results:

• MIX-TPI demonstrated superior performance compared to existing state-of-the-art methods.
• The framework showed strong generalization capabilities on independent datasets.
• Achieved effective prediction of TCR-pMHC interactions using sequence and property data.

Conclusions:

• MIX-TPI provides an effective computational solution for predicting TCR-pMHC interactions.
• The method addresses limitations of experimental approaches.
• The developed framework advances the field of immunoinformatics.