Abstract

Spatial omics (SO) is a powerful methodology that enables the study of genes, proteins, and other molecular features within the spatial context of tissue architecture. With the growing availability of SO datasets, researchers are eager to extract biological insights from larger datasets for a more comprehensive understanding. However, existing approaches focus on batch effect correction, often neglecting complex biological patterns in tissue slices, complicating feature integration and posing challenges when combining transcriptomics with other omics layers. Here, we introduce stMSA (SpaTial Multi-Slice/omics Analysis), a deep graph contrastive learning model that incorporates graph auto-encoder techniques. stMSA is specifically designed to produce batch-corrected representations while retaining the distinct spatial patterns within each slice, considering both intra- and inter-batch relationships during integration. Extensive evaluations show that stMSA outperforms state-of-the-art methods in distinguishing tissue structures across diverse slices, even when faced with varying experimental protocols and sequencing technologies. Furthermore, stMSA effectively deciphers complex developmental trajectories by integrating spatial proteomics and transcriptomics data, and excels in cross-slice matching and alignment for 3D tissue reconstruction.