SWIMR framework improves cell-type deconvolution in spatial transcriptomics by integrating multiple references
Researchers introduce a spatially weighted method that draws on several single-cell RNA-sequencing reference panels simultaneously, addressing accuracy limitations when tissues are complex or reference data are incomplete.
A methods paper published in PLOS Genetics on 2 November 2026 presents SWIMR (Spatially Weighted Integration of Multiple References), a computational framework for deconvolving cell-type composition from spatial transcriptomics data.
Spatial transcriptomics captures gene expression across tissue sections while preserving information about where in the tissue each measurement was made. However, many spatial technologies capture expression from areas that contain a mixture of cell types rather than from single cells. Determining what proportion of each cell type is present at each location — a task called deconvolution — is therefore essential for downstream analyses. Existing deconvolution methods typically rely on a single reference panel derived from single-cell RNA sequencing (scRNA-seq), which can introduce inaccuracies when the reference does not fully represent the cell types present in the tissue being analysed.
SWIMR, developed by Peiyao Zhao, Justin A. Colacino, Ying Ma, and Xiang Zhou of the University of Michigan, integrates information from multiple scRNA-seq reference panels while also using the spatial arrangement of measurements to improve estimates. Benchmarking in the paper shows that SWIMR outperforms single-reference methods in complex tissue settings, particularly when reference datasets are incomplete.
The method is applicable to studies of any tissue where spatial context is biologically important, and has implications for research into tumour microenvironments, developmental biology, and tissue homeostasis.
Sources
Read the original reporting — these are the public sources this summary draws from.
-
Primary source Public Library of Science · 2026-11-02Enhancing cell type deconvolution in spatial transcriptomics through incorporating multiple reference panels