MIST: An interpretable and flexible deep learning framework for single-T cell transcriptome and receptor analysis.

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Single-cell transcriptomic and T cell receptor (TCR) analyses are crucial for understanding T cell immunity.
• Integrating these data types offers deeper insights into T cell function and antigen specificity.
• Existing methods may lack the flexibility or interpretability for comprehensive T cell analysis.

Purpose of the Study:

• To introduce MIST (Multi-insight for T cell), a novel deep learning framework for joint analysis of single-cell transcriptomics and TCR data.
• To demonstrate MIST's capability in resolving T cell function and antigen specificity.
• To explore T cell heterogeneity in lung cancer immunotherapy and COVID-19 contexts.

Main Methods:

• Development of a deep learning framework, MIST, incorporating gene expression, TCR, and joint latent spaces.
• Vectorization and integration of single-cell transcriptome and TCR data.
• Application of MIST to datasets including antigen-specific T cells, lung cancer immunotherapy, and COVID-19.

Main Results:

• MIST accurately resolves T cell function and antigen specificity by integrating transcriptomic and TCR data.
• Analysis of lung cancer immunotherapy data revealed heterogeneity within CXCL13+ CD8+ T cell subsets.
• MIST provided novel insights into the functional transitions of CXCL13+ T cells during anti-PD-1 therapy.

Conclusions:

• MIST is an interpretable and flexible deep learning tool for joint single-cell transcriptomic and TCR analysis.
• The framework enhances the understanding of T cell heterogeneity and function in various immunological contexts.
• MIST offers valuable insights into immune responses, particularly in cancer immunotherapy and infectious diseases.