spCLUE: a contrastive learning approach to unified spatial transcriptomics analysis across single-slice and multi-slice data

Abstract

Advances in spatial transcriptomics demand new tools to integrate data across tissue slices and identify consistent spatial domains. We introduce spCLUE, a comprehensive framework combining multi-view graph network, contrastive learning, attention mechanisms, and a batch prompting module to learn informative spot representations and integrate data from both aligned and unaligned samples. spCLUE outperforms nine single-slice and seven multi-slice methods when tested on diverse datasets and reveals biologically relevant domains across different tissues and conditions. spCLUE offers a powerful solution to spatial domain analysis and integration in spatial transcriptomics, enabling more accurate and interpretable studies of tissue organization.