Spatial Metagene Discovery and Associated Molecular Pattern Characterization in Spatial Transcriptomics and Multi-Omics using SEPAR

Abstract

Spatially resolved transcriptomics (SRT) has transformed biomedical research by enabling gene expression profiling at near- or sub-cellular resolution while preserving spatial context. However, interpreting SRT data to understand cellular and gene organization remains challenging. Current methods focus on identifying global spatial domains across all genes, but often miss localized structures driven by specific gene subsets. Here, we introduce SEPAR, an unsupervised computational framework designed to analyze SRT data using a limited number of spatial metagenes and their expression patterns. SEPAR integrates gene activity and spatial neighborhood relationships to summarize gene expression per spot/- cell, decomposing data into contributions from spatial metagenes. It supports downstream analyses, including metagene expression pattern-specific gene identification, spatially variable gene (SVG) detection, spatial domain delineation, and gene expression refinement. Applied to diverse datasets, SEPAR demonstrates high efficacy. Gene sets linked to spatial metagene patterns are enriched with meaningful cell types and gene ontologies. SVGs are detected with higher accuracy, and gene refinement enhances biological signals, improving correlation analysis of functionally related genes. In spatial multi-omics data, SEPAR excels at revealing strong correlations between co-localized molecules in spatial CITE-seq data and identifies coordinated gene-peak relationships in MISAR-seq data, offering insights into spatial molecular interactions.