deconvolution-cell2location

Bayesian deconvolution estimates spatial transcriptomics cell-type proportions by using a two-stage model that first learns cell type signatures from a scRNA-seq reference and then maps them to spatial spots to calculate per-spot composition.

The best way to deconvolve Visium data is using a two-stage Bayesian model that learns cell type signatures from an annotated scRNA-seq reference and maps them to spatial locations, outputting per-spot proportions and dominant cell types.

Yes, you need a GPU for efficient training when running cell2location to deconvolve single-slice spatial transcriptomics datasets to estimate cell-type proportions and provide uncertainty estimates.

Bayesian deconvolution quantifies uncertainty by providing credible intervals for each cell type proportion estimated per spatial spot, allowing you to assess the confidence of the mapped cell-type composition across tissue sections.

You need an annotated single-cell RNA-seq reference file with a celltype label column and a spatial spot dataset, which the model uses to output deconvolution weights and dominant cell types per spot.

Deconvolution results are stored in adata.obsm['deconv_weights'] for per-spot cell-type proportions and adata.obs['celltype'] for the dominant cell type, enabling seamless downstream analysis integration.