STP: single-cell partition for subcellular spatially-resolved transcriptomics

Spatially-resolved transcriptomics (SRT) technologies now allow exploration of gene expression with spatial context. Recent advances achieving subcellular resolution provide richer data but also introduce challenges, such as aggregating subcellular spots into individual cells, which is a task distinct from traditional deconvolution. Existing methods often grid SRT data into predefined squares, which is unrealistic for accurately capturing cellular boundaries. We propose a method, STP, that integrates subcellular SRT data with nuclei-stained images to partition individual cells. STP first segments nuclei and maps their masks onto the SRT data, then uses a simulated-annealing-inspired approach to expand nuclear boundaries to the full cellular level. Evaluated on subcellular SRT datasets from Drosophila embryos at multiple developmental stages and from mouse embryos with a large field-of-view, STP demonstrated accurate single-cell partitioning, unveiling significant spatial tissue patterns and identifying undetected cell types beyond previous methods. Existing methods to analyse spatially-resolved transcriptomics (SRT) data often fail to accurately capture cell boundaries. The authors propose STP, a method that integrates subcellular SRT data with nuclei-stained images to partition individual cells. STP demonstrates effective cell partitioning when applied to SRT data from fly and mouse embryos.