Rapid and memory-efficient analysis and quality control of large spatial transcriptomics datasets

Abstract

The 10x Visium spatial transcriptomics platform has been widely adopted due to its established analysis pipelines, robust community support, and manageable data output. However, technologies like 10x Visium have the limitation of being low-resolution, and recently spatial transcriptomics platforms with subcellular resolution have proliferated. Such high-resolution datasets pose significant computational challenges for data analysis, with regards to memory requirement and processing speed. Here, we introduce Pseudovisium, a Python-based framework designed to facilitate the rapid and memory-efficient analysis, quality control and interoperability of high-resolution spatial transcriptomics data. This is achieved by mimicking the structure of 10x Visium through hexagonal binning of transcripts. Analysis of 47 publicly available datasets concluded that Pseudovisium increased data processing speed and reduced dataset size by more than an order of magnitude. At the same time, it preserved key biological signatures, such as spatially variable genes, enriched gene sets, cell populations, and gene-gene correlations. The Pseudovisium framework allows accurate simulation of Visium experiments, facilitating comparisons between technologies and guiding experimental design. Specifically, we found high concordance between Pseudovisium (derived from Xenium or CosMx) and Visium data from consecutive tissue slices. We further demonstrate Pseudovisium's utility by performing rapid quality control on large-scale datasets from Xenium, CosMx, and MERSCOPE platforms, identifying similar replicates, as well as potentially low-quality samples and probes. The common data format provided by Pseudovisium also enabled direct comparison of metrics across 6 spatial transcriptomics platforms and 59 datasets, revealing differences in transcript capture efficiency and quality. Lastly, Pseudovisium allows merging of datasets for joint analysis, as demonstrated by the identification of shared cell clusters and enriched gene sets in the mouse brain using data from multiple spatial platforms. By lowering the computational requirements and enhancing interoperability and reusability of spatial transcriptomics data, Pseudovisium democratizes analysis for wet-lab scientists and enables novel biological insights.