TransTCR: Integrating TCRs and Transcriptomes Through Optimal Transport for Antigen Specificity Prediction

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Single-cell immune profiling allows simultaneous measurement of T cell receptors (TCRs) and transcriptomes.
• Heterogeneity between TCR and transcriptome data presents challenges for effective integration in adaptive immunity studies.
• Existing methods may overlook modality-specific biological signals due to simplistic alignment assumptions.

Purpose of the Study:

• To introduce TransTCR, a novel multimodal learning framework for integrating TCR and transcriptomic data.
• To address the challenge of aligning heterogeneous multimodal T cell data.
• To improve the characterization of adaptive immunity at single-cell resolution.

Main Methods:

• TransTCR utilizes optimal transport (OT) and contrastive learning to align TCR and transcriptomic representations.
• Pretrained foundation models extract features from each modality.
• OT-based projection into a shared latent space achieves distribution-aware alignment, refined by a bidirectional contrastive objective.

Main Results:

• TransTCR significantly outperforms existing single-modality and multimodal baselines.
• Achieved state-of-the-art performance in antigen specificity recognition tasks.
• Demonstrated superior results in intra-donor and inter-donor antigen specificity prediction and clustering.

Conclusions:

• TransTCR is a powerful computational tool for integrating and analyzing multimodal T cell data.
• Facilitates deeper insights into adaptive immune responses.
• Enhances the understanding of TCR-transcriptome relationships in immunology.