STRESS: spatial transcriptomics resolution enhancing method based on the state-space model.

The widespread application of spatially resolved transcriptomics (SRT) has provided a wealth of data for characterizing gene expression patterns within the spatial microenvironments of various tissues. However, the relatively coarse spatial resolution of most SRT platforms limits the continuity and interpretability of spatial expression landscapes, particularly for downstream analysis such as spatial domain identification. To address this limitation, we present STRESS, a deep learning framework for tissue-level spatial expression refinement using only SRT gene expression profiles and spatial coordinates, without relying on histological images or single-cell references. STRESS adopts a 3D state-space modeling architecture to jointly capture spatial dependencies among neighboring locations and transcriptional relationships across genes, enabling the estimation of spatially coherent expression patterns on finer spatial grids. We evaluate STRESS across multiple datasets spanning different platforms and tissue types, and demonstrate that the refined spatial representations consistently enhance spatial domain delineation and stability in downstream analyses. These results highlight STRESS as a practical and reference-free approach for refining tissue-scale spatial transcriptomic patterns and facilitating integrative spatial data analysis.