Speaker: 张雷，同济大学

Title: vqTCR: A Product-Quantized Representation Framework Linking TCR Specificity to Transcriptomic Cell States

Language: Chinese

Time & Venue: 2025.12.02 14:00-15:00 思源楼 315会议室

Abstract: T cell heterogeneity arises from both antigen-specific TCRs and transcriptional regulation, but linking receptor specificity to functional states across individuals remains challenging. We introduce vqTCR, a product vector-quantized variational auto-encoder that learns discrete, combinatorial representations linking TCR sequences to transcriptional programs. vqTCR decomposes the latent space into interpretable T cell prototypes, capturing receptor-function relationships while maintaining flexibility. A prototypical baseline module combined with FiLM-modulated residuals disentangles prototypical functional states from context-specific variations. Evaluations show that vqTCR enables cross-sample and cross-clonotype antigen-binding prediction, functional state visualization, TCR-gene expression association analysis, and reveals biologically consistent gene-level feature modulation. Our framework provides an interpretable approach for multi-modal immune representation learning, elucidating how TCR specificity couples with cellular state in adaptive immunity.