TSTScope reveals T cell phenotypes predictive of immune checkpoint blockade response.

TSTScope is a multimodal deep learning framework that jointly integrates scRNA-seq and scTCR-seq data. TSTScope utilizes a gene program (GP)-constrained variational autoencoder (VAE) to construct an interpretable latent space, effectively unifying antigen specificity with transcriptional states. The datasets for this study is being prepared for public release and will be continuously uploaded during the submission process.