SEPAR enables spatial metagene discovery and associated molecular pattern characterization in spatial transcriptomics and multi-omics datasets.

Spatially resolved transcriptomics (SRT) profiles gene expressions at near- or sub-cellular resolution while preserving their spatial context, yet interpreting SRT data to understand spatial cellular and molecular organization remains challenging. Most existing computational methods focus on global spatial domains but overlook localized structures driven by specific gene subsets. Here, we introduce SEPAR, an unsupervised framework that leverages spatial metagenes to analyze SRT data by integrating gene activity and spatial neighborhood relationships. It enables multiple downstream analyses including: identifying metagene pattern-specific genes, detecting spatially variable genes (SVGs), delineating spatial domains, and refining expression signals. Evaluated on diverse datasets, SEPAR reveals biologically meaningful gene ontologies and cell types in gene sets linked to metagene patterns, identifies SVGs with higher accuracy, and enhances biological signals with gene refinement. In spatial multi-omics data, it uncovers co-localized molecule correlations in spatial CITE-seq and coordinated gene-peak relationships in MISAR-seq, offering insights into spatial molecular interactions.