This project analyzes Visium HD spatial transcriptomics data from colorectal cancer (CRC) samples using a deep learning framework. Visium HD delivers unprecedented spatial resolution (16 µm bins), enabling near-single-cell analysis of the tumor microenvironment. The massive dataset size (>140,000 spatial locations per sample, >260,000 reference single cells) creates computational bottlenecks that exceed standard workstation capabilities. We will use a Bayesian deep learning model to decompose mixed gene expression signals into contributions from distinct cell types. This requires GPU acceleration for efficient variational inference. The project maps the spatial distribution of tumor cells, immune populations, and stromal components, providing insights into tumor-immune interactions for the development of therapeutic strategies.