deconvolution-flashdeconv

Spatial transcriptomics deconvolution estimates per-spot cell type proportions by applying a reference-based method to your gene-expression data. This Skill uses an annotated scRNA-seq reference to return per-spot proportions and a dominant cell type for each spot.

No GPU is needed to run reference-based spot deconvolution on large spatial datasets. This Skill provides a Python-based workflow using the FlashDeconv method, designed to be ultra-fast and scalable for single-slice gene-expression data on standard compute.

Reference-based deconvolution requires an .h5ad file containing the scRNA-seq reference dataset. The reference must include a cell type column, with the default column name expected to be 'celltype', to successfully map cell proportions to spatial spots.

Deconvolution results are stored directly within the anndata object. The estimated per-spot cell type proportions are saved in adata.obsm['flashdeconv'], while the dominant cell type identified for each individual spot is stored in adata.obs['celltype'].

The best way to deconvolve a single-slice spatial transcriptomics dataset without GPU acceleration is using the FlashDeconv method. It performs ultra-fast, reference-based estimation of cell-type composition per spot using Python, scanpy, pandas, and numpy.

No, spatial transcriptomics deconvolution with this Skill requires a properly formatted scRNA-seq reference containing a celltype column. Without annotated cell types in the reference .h5ad file, the algorithm cannot map reference profiles to estimate per-spot proportions.