CAESAR: a cross-technology and cross-resolution framework for spatial omics annotation

The biotechnology of spatial omics has advanced rapidly over the past few years, with enhancements in both throughput and resolution. However, existing annotation pipelines in spatial omics predominantly rely on clustering methods and lack the flexibility to integrate extensive annotated information from single-cell RNA sequencing (scRNA-seq) due to discrepancies in spatial resolutions, species, or modalities. Here we introduce the CAESAR suite, an open-source software package that provides image-based spatial co-embedding of locations and genomic features. It uniquely transfers labels from scRNA-seq reference data, enabling the annotation of spatial omics datasets across different technologies, resolutions, species, and modalities, based on the conserved relationship between signature genes and cells/locations at an appropriate level of granularity. Notably, CAESAR enriches for location-level pathways, allowing for the detection of gradual biological pathway activation within spatially defined domain types. We demonstrate the advantages of CAESAR through a comprehensive analysis of five spatial omics datasets encompassing diverse technologies, resolutions, and modalities. Across these applications, CAESAR achieved substantial improvements in annotation accuracy (45.45%-4333.33%) by transferring cell-type labels from either multiple reference data, or across different species and modalities. As a result, CAESAR effectively recovers intricate structures in mouse olfactory bulb and embryo, and unveils tumor microenvironment heterogeneity, with exceptional efficiency and flexibility.