Sainsc: a computational tool for segmentation-free analysis ofin-situcapture

Abstract

Spatially resolved transcriptomics has become the method of choice to characterise the complexity of biomedical tissue samples. Until recently, scientists have been restricted to profiling methods with high spatial resolution but for a limited set of genes or methods that can profile transcriptome-wide but at low spatial resolution. Through recent developments, there are now methods which offer subcellular spatial resolution and full transcriptome coverage. However, utilizing the high spatial and gene resolution of these new methods remains elusive due to several factors including low detection efficiency, high computational cost and difficulties in delineating cell borders. Here we present Sainsc (Segmentation-free analysis ofin-situcapture data), which combines a cell-segmentation free approach with efficient data processing of transcriptome-wide nanometer resolution spatial data. Sainsc can generate cell-type maps with accurate cell-type assignment at a subcellular level, together with corresponding maps of the assignment scores that facilitate the interpretation in the local confidence of cell-type assignment. We demonstrate its utility and accuracy across different tissues and profiling methods. Compared to other methods, Sainsc requires lower computational resources and has scalable performance, enabling interactive data exploration. Sainsc is compatible with common data analysis frameworks and is available as open-source software in multiple programming languages.