STCS: A UNIVERSAL FRAMEWORK FOR CELL SEGMENTATION IN SEQUENCING-BASED SPATIAL TRANSCRIPTOMICS WITH BENCHMARKING AGAINST CURRENT APPROACHES

Sequencing-based spatial transcriptomics now reaches subcellular resolution—Visium HD and Stereo-seq both capture transcriptome-wide signal at micrometer or even nanometer scale. Yet the raw output is a grid of barcoded bins, not cells. For any analysis that treats cells as the basic unit, this mismatch is a serious obstacle. Extract- ing coherent single-cell profiles from dense spatial bins remains difficult, and existing tools often fall short. STCS (Spatial Transcriptomics Cell Segmentation) tackles this problem as a univer- sal framework rather than a platform-specific pipeline. It starts from nuclei segmented in the histology image, then assigns surrounding bins to nuclei using a distance that blends transcriptomic similarity with spatial proximity. Only two parameters control this process: a search radius and a spatial weight. Instead of requiring matched ground truth, STCS chooses these parameters by tracking internal stability and spatial coher- ence, making it adaptable to different tissues and technologies. We benchmarked STCS against current approaches in two settings. First, on a Visium HD human lung cancer sample with matched Xenium annotations, we performed direct cell-level comparison. Second, on high-resolution Stereo-seq mouse brain data, we tested cross-platform generalizability. Across cell-type agreement, spatial organization, gene-expression fidelity, and compositional accuracy, STCS consistently matched or exceeded existing methods while keeping reconstructed cells spatially coherent. As sequencing-based spatial profiling spreads through biomedical research, STCS of- fers an open-source and broadly usable route from bins to cells, supporting downstream analyses that depend on reliable cell-level data.