SpaBatch: Batch Alignment of Spatial Transcriptomics Data using Graph Deep Learning

Abstract

Spatial transcriptomics (ST) has emerged as a transformative technology, enabling the simultaneous capture of gene expression and spatial context within tissues. However, batch effects arising from technical variability remain a significant barrier to integrating datasets generated from different experiments. To this end, we introduce a novel graph deep learning framework (SpaBatch) tailored for the batch alignment of multi-slice spatial transcriptomics data. SpaBatch is an innovative computational framework leveraging Variational Graph Autoencoders, self-supervised learning, and triplet learning with readout aggregation to enhance multi-slice spatial transcriptomics data integration. We validated our framework on multiple datasets encompassing various tissue types and experimental conditions. Through the adjustment of batch effects, SpaBatch facilitates the analysis of spatial transcriptome domains across multiple slices, showcasing its capability to reveal new biological insights in multi-slice spatial transcriptome studies.