Spatial RNA velocity reveals cellular state transitions and prognostic markers in the melanoma microenvironment.

The inability to infer transcriptional dynamics from high-resolution spatial transcriptomics represents a major computational challenge, as these datasets lack the spliced/unspliced mRNA counts required for conventional RNA velocity. To address this, we developed a novel computational framework that repurposes subcellular transcript localization-using nuclear and cytoplasmic RNAs as proxies for unspliced and spliced mRNA, respectively-for spatial RNA velocity analysis. By integrating this approach with scVelo's dynamical model, we inferred directional state transitions directly from melanoma spatial transcriptomes. Our framework successfully reconstructed progression trajectories of melanoma cells and differentiation paths of infiltrating T cells, identifying cluster-specific dynamic genes. These genes were significantly associated with patient prognosis and formed protein-protein interaction networks enriched for immune-related pathways. This study provides a generalizable computational strategy to decode spatiotemporal dynamics from static spatial transcriptomic data, bridging a critical gap between spatial biology and transcriptional dynamics.