A systematic benchmark of batch effect correction methods for spatial transcriptomics

Spatial transcriptomics enables high-resolution profiling of gene expression within tissue slices, but its reliability is often compromised by technical batch effects that obscure biological signals and hinder data integration. A systematic approach to define, evaluate, and correct these artifacts is critically needed. Here, we establish SpaBEAT (Spatial Batch Effect Assessment and Testing), a systematic framework that defines four key types of batch effects in spatial transcriptomics: inter-slice, inter-sample, cross-protocol/platform, and intra-slice. Using this framework, we benchmark ten spatial integration methods across diverse spatial transcriptomics modalities, including spot-based, high-resolution, image-based targeted, and cross-platform datasets. We further introduce controlled and semi-synthetic simulations to disentangle technical variation from predefined biological differences, and evaluate method robustness to preprocessing choices, targeted-gene overlap, and cell-segmentation strategy. Performance is rigorously quantified using a panel of metrics for batch-effect removal and biological signal preservation, together with hierarchical ranking, task coverage and computational efficiency. Our analysis reveals that spatial batch-correction performance is context-dependent, with distinct trade-offs between batch-effect removal and the preservation of biological structure, and that no method is universally optimal across tissues, platforms, and batch-effect scenarios. Our work establishes a systematic and standardized framework for defining and assessing batch effects in spatial transcriptomics. SpaBEAT provides practical guidance for method selection and offers benchmark datasets, simulations, reproducible workflows, and evaluation resources to facilitate more robust and reproducible spatial transcriptomics research.