Integrating spatial transcriptomics count data with Crescendo improves visualization and detection of spatial gene patterns

Spatial transcriptomics allows for the analysis of a cell’s gene expression in the context of its physical location. With spatial transcriptomics data, investigators often want to find genes of interest whose spatial patterns are biologically relevant in multiple samples. However, due to confounding factors in spatial data that produce noise across samples, datasets, and technologies, it is challenging to visualize genes and their spatial patterns across samples. We present Crescendo, an integration algorithm that performs correction directly on gene expression counts to reduce variation from technical confounders. We first apply Crescendo to a 3-sample spatial transcriptomics mouse brain dataset to show how Crescendo enables accurate visualization of gene expression across these spatial transcriptomic samples. We then demonstrate Crescendo’s scalability by integrating a 16-sample immuno-oncology dataset of 7 million cells. Finally, we show that Crescendo can perform cross-technology integration by merging a colorectal cancer (CRC) scRNA-seq dataset with two CRC spatial transcriptomics samples. By transferring information between technologies, Crescendo can impute poorly expressed genes to improve detection of gene-gene colocalization, such as ligand-receptor interactions.