Deciphering Spatial Cellular Organization through RNA-Anchored Multi-Omics Deconvolution

Abstract

Spatial technologies that jointly profile RNA and additional molecular layers provide a comprehensive view of tissue organization, yet their limited resolution results in measurements that reflect mixtures of multiple cell types. While RNA-based deconvolution methods are well established, incorporating co-profiled non-transcriptomic modalities without compromising cell-type identifiability remains an open challenge, as reliable references are largely restricted to transcriptomic data. Here we present SMODER, an RNA-anchored framework for spatial multi-omics deconvolution that treats transcriptomes as the primary determinant of cell identity while incorporating complementary molecular modalities in a constrained manner. By adopting an asymmetric integration strategy, SMODER preserves RNA-driven cell-type resolution and enables robust estimation of cell-type compositions across spatial locations. Across simulations and diverse spatial multi-omics datasets spanning protein, chromatin accessibility and histone modification assays, SMODER consistently outperforms existing methods in both accuracy and spatial coherence. Beyond deconvolution, SMODER denoises multi-omic signals and enables cell-type-aware analysis of regulatory programs. Application to spatial CUT&Tag–RNA-seq and RNA–ATAC-seq data reveals spatially resolved epigenetic repression and cis-regulatory relationships linked to gene expression.