fuzzyCell

Cell type labels in spatial single-cell data are uncertain, particularly between closely related subtypes (for example CD4+ and CD8+ T cells), but most downstream analyses treat them as known. fuzzyCell estimates a per-cell classification confidence by bootstrapping (resampling each cell's transcript counts and re-classifying it), and uses the confidences as soft cell type weights in downstream spatial analyses:

• cell type proportions, among all cells and relative to the parent group;
• co-localisation of pairs of cell types, with spicyR's L-function, and relative co-localisation with Statial's Kontextual test (via spicyR);
• niche identification, from lisaClust's local indicators of spatial association.

Each analysis is run with the original hard labels and with the confidence weights, so the two can be compared, and each can be tested against a sample-level outcome. Cell types are organised with a user-supplied hierarchy of lineages, groups and subtypes.

Installation

Development version

The development version is on GitHub. It depends on several Bioconductor packages (for example SpatialExperiment and Statial), so install it with BiocManager, which installs those dependencies from Bioconductor:

if (!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager")
BiocManager::install("shreyarajeshrao/fuzzyCell")

Or, with remotes, which also needs the Bioconductor repositories set:

# install.packages("remotes")
remotes::install_github("shreyarajeshrao/fuzzyCell",
    repos = BiocManager::repositories())

fuzzyCell requires R >= 4.5.0. To also build the vignette, add build_vignettes = TRUE. This needs the Quarto command line tool, the quarto R package, and the example data package SpatialDatasets:

BiocManager::install(c("quarto", "SpatialDatasets"))
BiocManager::install("shreyarajeshrao/fuzzyCell", build_vignettes = TRUE)

Bioconductor release

fuzzyCell is being prepared for Bioconductor. Once it is accepted it will be installed with:

BiocManager::install("fuzzyCell")

Quick start

library(fuzzyCell)

# heart transplant Xenium biopsies (downloads the data on first use)
spe = loadAmancherla("pre")
data(heartCellTypeTree)

# per-cell confidence for each group of related subtypes
confList = bootstrapLineageConfidence(spe, heartCellTypeTree, nBoot = 50)

# weighted vs unweighted analyses, tested against the outcome
props = runProportionsPipeline(spe, heartCellTypeTree, confList,
    outcomeCol = "resolution_broad", subject = "subject")
coloc = runColocalisationModels(spe, heartCellTypeTree, confList,
    outcomeCol = "resolution_broad", subject = "subject")

# the weights themselves, to pass to spicyR or lisaClust directly
w = confidenceWeights(spe, heartCellTypeTree, confList)

The input is any SummarizedExperiment (such as a SpatialExperiment) with a "counts" assay and cellID, imageID, cellType, x and y in its colData.

See vignette("fuzzyCell") for a full walkthrough.

Issues

Please report bugs and requests at https://github.com/shreyarajeshrao/fuzzyCell/issues.

Development

Parts of fuzzyCell (package code, tests, documentation and the vignette) were written with the assistance of Claude (Anthropic), via Claude Code, in line with Bioconductor's AI and third-party code policy. All code has been reviewed and tested by the maintainer, who is responsible for its correctness and maintenance.