FineST: contrastive learning integrates histology and spatial transcriptomics for nuclei-resolved ligand-receptor analysis - PubMed
3 hours ago
- #cell-cell communication
- #spatial transcriptomics
- #contrastive learning
- FineST is a deep contrastive learning model integrating histology and spatial transcriptomics (ST) for nuclei-resolved ligand-receptor analysis.
- It addresses limitations of ST, such as low resolution and data sparsity, by leveraging histology images for precise nuclei segmentation and high-resolution RNA imputation.
- FineST outperforms existing methods in RNA imputation, cell type prediction, and cell-cell communication (CCC) pattern discovery.
- Applied to colorectal and breast cancer datasets, it reveals novel tumor-immune interactions, including invasive fronts, tertiary lymphoid structures, and therapy resistance barriers.
- The study highlights a new paradigm in ST analysis by integrating histology images, enhancing CCC insights across cancer types.