Package 'fuzzyCell'

Description

The single-group worker behind bootstrapLineageConfidence(). Trains a classifier on one group's cells, then repeatedly resamples each cell's counts (multinomial resampling) and re-predicts its class. The fraction of resamples assigning a cell to each class is its confidence for that class.

Usage

bootstrapClassify(
  countMat,
  labels,
  nBoot = 100,
  nPerClass = 50,
  cores = 1,
  seed = NULL,
  classifier = c("pcaLogistic", "logistic", "rf", "lda", "svm"),
  ntree = 500,
  priors = c("balanced", "observed"),
  correctedCounts = FALSE
)

Arguments

Value

A numeric matrix with one row per cell (in the order of countMat) and one column per class in labels. Entry ⁠[i, k]⁠ is the fraction of bootstrap resamples in which cell i was classified as class k, so each row sums to 1.

Examples

data(lungSubset)
endo = lungSubset$cellType %in% c("Arteriole", "Venous", "Capillary", "Lymphatic")
countMat = t(as.matrix(SummarizedExperiment::assay(lungSubset, "counts")[, endo]))
freq = bootstrapClassify(countMat, lungSubset$cellType[endo], nBoot = 5, seed = 1)
head(freq)

Description

For each group in cellTypeTree that has children (see cellTypeTree), runs bootstrapClassify() on that group's cells to get, for every cell, a confidence score for each of the group's child cell types. The result is the confList used by the downstream analyses, such as runProportionsPipeline() and runColocalisationModels(). Cell types without a group (ungrouped leaves) are not refined and keep hard labels.

Usage

bootstrapLineageConfidence(
  spe,
  cellTypeTree,
  nBoot = 100,
  nPerClass = 50,
  cores = 1,
  seed = 51773,
  skipGroups = character(0),
  subprocess = FALSE,
  classifier = c("pcaLogistic", "logistic", "rf", "lda", "svm"),
  ntree = 500,
  priors = c("balanced", "observed"),
  correctedCounts = FALSE,
  cellType = "cellType",
  cellID = "cellID",
  assayName = "counts"
)

Arguments

Value

A named list with one element per group (named by group). Groups with no cells, or with only one cell type present, cannot be bootstrapped: they are skipped with a message and left out of the list. Each element is a data.frame with one row per cell in that group, a cellID column, and one column per child cell type of the group holding the cell's confidence score for that cell type (between 0 and 1, rows summing to 1 across the child columns). Each data frame also carries a "settings" attribute recording how it was computed (nBoot, nPerClass, classifier, priors, correctedCounts, ntree, the group's seed, its children and the fuzzyCellVersion).

Examples

data(lungSubset)
data(lungCellTypeTree)
confList = bootstrapLineageConfidence(lungSubset, lungCellTypeTree, nBoot = 5,
    seed = 1)
names(confList)
head(confList[["Endothelial lineage"]])

# the same call with nBoot = 20 gives the confidences in `lungConfList`

Description

Helpers for the cell type tree: a nested list describing how cell types are organised into broad lineages and (optionally) groups of closely related subtypes. bootstrapLineageConfidence() refines the cells within each group, and the analysis functions compare cell types, groups and lineages.

Usage

buildCellTypeMap(cellTypeTree)

annotateCellTypeTree(spe, cellTypeTree, cellType = "cellType")

resolveTreeLeaves(cellTypeTree, nodeName)

autoParentOverrides(cellTypeTree)

Arguments

Details

A tree is a named list with one entry per broad lineage (these become the lineage column). Each lineage is either a character vector of cell types, or a list whose entries are:

• a bare string: a cell type that is not in any group;

• a named list(children = c(...)): a group of closely related cell types, the unit that is bootstrap-refined. A group may contain only children;

• a named list of further entries: purely organisational nesting (a "wrapper") that is not a group itself, to any depth.

A lineage can also itself be list(children = c(...)), in which case the lineage name doubles as the group name. A cell type's group is its nearest list(children = ...) ancestor, however many wrappers sit above it. In the example below CD8_Resting is in the group ⁠CD8+ T cells⁠, not ⁠T cells⁠.

Every cell type must appear exactly once, and lineage, wrapper and group names must be unique and different from every cell type name.

list(
    "Immune" = list(
        "T cells" = list(
            "CD8+ T cells" = list(children = c("CD8_Resting", "CD8_Memory")),
            "CD4+ T cells" = list(children = c("CD4_naive", "CD4_Memory"))),
        "NK",
        "Mast"),
    "Endothelial lineage" = c("Endothelial", "Activated endothelial"))

resolveTreeLeaves() expands any node, however far up the tree, to the cell types beneath it. It is used by relative co-localisation when it conditions on a parent that is broader than a cell type's own group (see parentOverrides in runColocalisationModels()), and autoParentOverrides() finds those parents automatically.

Value

• buildCellTypeMap(): a list with cellMap (a data.frame with columns cellType, group and broad, one row per cell type) and groupsWithChildren (names of the groups to be bootstrap-refined). An error is raised if the tree is not a valid tree as described in Details: it must be a non-empty named list, every top-level entry must be named, groups may only contain children, cell type names must be non-empty strings, and names must not be repeated or shared between cell types and lineages, wrappers or groups.

• annotateCellTypeTree(): spe with group and lineage columns added (a message reports any existing group or lineage column whose values are replaced). An error is raised if any cell's cell type is not in the tree (including NA).

• resolveTreeLeaves(): character vector of every cell type beneath nodeName.

• autoParentOverrides(): named character vector mapping each group that has a named ancestor to that ancestor, for use as parentOverrides in runColocalisationModels().

Examples

data(lungCellTypeTree)
buildCellTypeMap(lungCellTypeTree)
resolveTreeLeaves(lungCellTypeTree, "Immune")
autoParentOverrides(lungCellTypeTree)

data(lungSubset)
cells = annotateCellTypeTree(getCellData(lungSubset), lungCellTypeTree)
table(cells$group, useNA = "ifany")

Description

Builds the soft cell type membership of every cell, the weights that the weighted analyses pass to spicyR and lisaClust: one column per leaf cell type and per group of the tree. A cell of a bootstrapped group has, for each subtype of the group, its bootstrap confidence in that subtype; every other cell has weight 1 for its own cell type. A group's column is 1 for each of its cells, since the bootstrap reclassifies cells only within their group.

Usage

confidenceWeights(
  spe,
  cellTypeTree,
  confList,
  weighted = TRUE,
  cellType = "cellType",
  cellID = "cellID"
)

Arguments

Value

A numeric matrix with one row per cell of spe (in its order) and one named column per leaf cell type and group.

Examples

data(lungSubset)
data(lungCellTypeTree)
data(lungConfList)
w = confidenceWeights(lungSubset, lungCellTypeTree, lungConfList)
head(w[, c("Arteriole", "Venous", "Endothelial lineage")])

Description

An alternative to bootstrapLineageConfidence(): for each group with children, a classifier is trained on the group's cells with balanced classes and gives each cell the probability of each child cell type. The cells are split into nFolds folds and each fold is scored by a classifier trained on the other folds, so no cell is scored by a classifier that saw it. Because the classes are balanced in training, the probabilities give every child cell type the same prior weight.

Usage

crossfitLineageProbabilities(
  spe,
  cellTypeTree,
  nFolds = 5,
  nPerClass = 50,
  seed = 51773,
  classifier = c("pcaLogistic", "logistic", "rf", "lda"),
  ntree = 500,
  priors = c("balanced", "observed"),
  cores = 1,
  skipGroups = character(0),
  cellType = "cellType",
  cellID = "cellID",
  assayName = "counts"
)

Arguments

Value

A confidence list in the format of bootstrapLineageConfidence(), with probabilities in place of bootstrap confidences.

Examples

data(lungSubset)
data(lungCellTypeTree)
probList = crossfitLineageProbabilities(lungSubset, lungCellTypeTree,
    nFolds = 3, seed = 1)
head(probList[["Endothelial lineage"]])

Description

Returns colData(spe) as a data.frame. For a SpatialExperiment whose coordinates are stored in spatialCoords() rather than in colData(), the coordinates are added as columns x and y. The spatial functions in fuzzyCell expect cellID, imageID, cellType, x and y columns.

Usage

getCellData(spe)

Arguments

Value

A data.frame with one row per cell.

Examples

data(lungSubset)
head(getCellData(lungSubset))

Description

The cell type hierarchy used with loadAmancherla(): broad lineages, groups of closely related subtypes (for example the T cell subtypes), and ungrouped leaf cell types. See cellTypeTree for the format.

Usage

data(heartCellTypeTree)

Format

A nested list; see cellTypeTree.

Source

Defined in data-raw/heartCellTypeTree.R.

Examples

data(heartCellTypeTree)
buildCellTypeMap(heartCellTypeTree)$groupsWithChildren

Description

Downloads the Amancherla et al. (2025) heart transplant Xenium dataset from the SpatialDatasets package (via ExperimentHub) and prepares it for fuzzyCell: renames the columns fuzzyCell expects, keeps patients with cellular rejection, and keeps one randomly chosen image per patient for each biopsy timing. The first call downloads the data, which is large. With "both" every patient has an image before and after transplant, so those images are not independent and cannot be passed together to runColocalisationModels() with subject = "subject".

Usage

loadAmancherla(timing = c("pre", "post", "both"), seed = 51773)

Arguments

Details

The prepared object has the columns fuzzyCell functions use: cellID, imageID, cellType, x, y and subject, and the per-patient outcome resolution_broad ("persistence" or "resolved"). The cell type hierarchy for these cell types is in heartCellTypeTree.

Value

A SpatialExperiment.

Examples

spe = loadAmancherla("pre")
spe
table(spe$resolution_broad)

Description

The cell type hierarchy for lungSubset (and the full SpatialDatasets::spe_Vannan_2025() data): broad lineages, groups of closely related subtypes (for example the fibroblast subtypes), and ungrouped leaf cell types. See cellTypeTree for the format.

Usage

data(lungCellTypeTree)

Format

A nested list; see cellTypeTree.

Source

Defined in data-raw/lungCellTypeTree.R.

Examples

data(lungCellTypeTree)
buildCellTypeMap(lungCellTypeTree)$groupsWithChildren

Description

The bootstrap classification confidences of the cells in lungSubset, from bootstrapLineageConfidence(lungSubset, lungCellTypeTree, nBoot = 20, classifier = "pcaLogistic", seed = 1), so that the examples can skip the bootstrap.

Usage

data(lungConfList)

Format

A named list with one data frame per group of lungCellTypeTree (see bootstrapLineageConfidence() for the format).

Source

Built by data-raw/lungConfList.R.

Examples

data(lungConfList)
names(lungConfList)
head(lungConfList[["Endothelial lineage"]])

Description

Xenium data of lung biopsies from Vannan et al. (2025), available in full as SpatialDatasets::spe_Vannan_2025(), cut down to ship with the package. It has five control and five pulmonary fibrosis biopsies (the five with the fewest cells of each), each cropped to the roughly 3,000 cells in a square window at its centre, so each image is a window of a larger biopsy. Every image comes from a different subject. Cell types with fewer than 20 cells in the windows are removed. The hierarchy for these cell types is in lungCellTypeTree.

Usage

data(lungSubset)

Format

A SpatialExperiment with the raw counts of 343 genes in 29,971 cells (assay counts), the cell coordinates in spatialCoords(), and in colData:

cellID

A unique cell ID.

imageID

The image (the window of a biopsy); also sample_id.

subject

The patient.

cellType

The cell type label.

disease_status

"Control" or "Disease".

Source

Built from SpatialDatasets::spe_Vannan_2025() by data-raw/lungSubset.R.

Examples

data(lungSubset)
lungSubset
table(lungSubset$imageID, lungSubset$disease_status)

Description

Defines niches by k-means clustering of per-cell feature vectors, once with hard cell type membership and once with bootstrap-confidence weights. The features of each cell are its local indicators of spatial association from lisaClust (lisaClust::lisa()) at several radii.

Usage

runNicheIdentification(
  spe,
  cellTypeTree,
  confList,
  toLabels = NULL,
  r = c(20, 50, 100),
  k,
  window = "convex",
  lisaFunc = "K",
  cores = 1,
  seed = 51773,
  cellType = "cellType",
  cellID = "cellID",
  imageID = "imageID",
  x = "x",
  y = "y"
)

Arguments

Details

For every cell, every label in toLabels and every radius in r, the feature is the local co-localisation of the cell's neighbours with that label (lisaClust's local K or L function, centred so that 0 means no attraction or repulsion). The weighted features count each neighbour with its bootstrap confidence of having the label, instead of 0 or 1 (see confidenceWeights()), and a label's density is its summed weight per unit area.

Niche names (niche1, niche2, ...) are arbitrary. K-means starts from random cluster centres, so the niches depend on the random seed. The unweighted and weighted clusterings are both started from the same seed, so that if the weights made no difference the two sets of niches would be identical.

Value

A list with the feature matrices (lisaMatUnweighted, lisaMatWeighted) and named cell-to-niche vectors nicheUnweighted and nicheWeighted.

Examples

data(lungSubset)
data(lungCellTypeTree)
data(lungConfList)
niches = runNicheIdentification(lungSubset, lungCellTypeTree, lungConfList,
    r = 50, k = 6)
table(weighted = niches$nicheWeighted, unweighted = niches$nicheUnweighted)

Description

plotNicheSpatial() shows the cells of one niche in one image. nicheCompositionMap() shows the relative enrichment of each cell type in each niche as a bubble plot or heatmap. nicheBarplot() shows it for a single niche. nicheProportionBoxplot() compares a niche's per-image proportion between outcome groups.

Usage

plotNicheSpatial(spe, imageID, niche, nicheCol = "nicheWeighted", cellSize = 1)

nicheCompositionMap(
  spe,
  niche,
  cellType = "cellType",
  type = c("bubble", "heatmap"),
  limit = c(0.33, 3),
  ...
)

nicheBarplot(spe, niche, target, cellType = "cellType", limit = c(0.33, 3))

nicheProportionBoxplot(proportions, niche, spe, outcomeCol)

Arguments

Value

A ggplot object; for nicheProportionBoxplot() a girafe (interactive) widget; for nicheCompositionMap(type = "heatmap") the result of pheatmap::pheatmap().

Examples

data(lungSubset)
data(lungCellTypeTree)
data(lungConfList)
niches = runNicheIdentification(lungSubset, lungCellTypeTree, lungConfList,
    r = 50, k = 6)
plotNicheSpatial(lungSubset, "VUHD113", "niche1",
    nicheCol = niches$nicheWeighted)
nicheCompositionMap(lungSubset, niches$nicheWeighted)
nicheBarplot(lungSubset, niches$nicheWeighted, "niche1")
res = testNicheOutcome(lungSubset, niches$nicheUnweighted,
    niches$nicheWeighted, outcomeCol = "disease_status", subject = "subject")
if (requireNamespace("ggiraph", quietly = TRUE)) {
    nicheProportionBoxplot(res$propWeighted, "niche1", lungSubset,
        "disease_status")
    nicheProportionBoxplot(res$propUnweighted, "niche1", lungSubset,
        "disease_status")
}

Description

Plots one image, with the cells of a group shaded by the (library-size normalised, rescaled) expression of one or two markers.

Usage

plotOneMarkerSpatial(
  spe,
  imageID,
  group,
  marker,
  colour = "purple",
  cellTypeTree = NULL,
  cellSize = 1,
  assayName = "counts"
)

plotTwoMarkerSpatial(
  spe,
  imageID,
  groupA,
  markerA,
  colourA = "purple",
  groupB,
  markerB,
  colourB = "green",
  cellTypeTree = NULL,
  cellSize = 1,
  assayName = "counts"
)

Arguments

Value

A ggplot object.

Examples

data(lungSubset)
data(lungCellTypeTree)
plotOneMarkerSpatial(lungSubset, "VUHD113", "Endothelial lineage", "HEY1",
    cellTypeTree = lungCellTypeTree)
plotTwoMarkerSpatial(lungSubset, "VUHD113", "Endothelial lineage", "PECAM1",
    "purple", "Fibroblast lineage", "COL1A1", "green",
    cellTypeTree = lungCellTypeTree)

Description

Shows, for one pair of cell types or groups, the per-image co-localisation measured with the original labels (unweighted) and with the bootstrap confidence weights (weighted), for each level of the outcome. Each point is one image.

Usage

plotMetricBoxplot(
  results,
  spe,
  fromType,
  toType,
  outcomeCol,
  metric = c("lfunction", "kontextual")
)

Arguments

Value

An interactive plot, a girafe widget that needs the ggiraph package.

Examples

data(lungSubset)
data(lungCellTypeTree)
data(lungConfList)
res = runColocalisationModels(lungSubset, lungCellTypeTree, lungConfList,
    outcomeCol = "disease_status", subject = "subject")
if (requireNamespace("ggiraph", quietly = TRUE)) {
    plotMetricBoxplot(res, lungSubset, "Mast", "Proliferating FBs",
        outcomeCol = "disease_status")
    plotMetricBoxplot(res, lungSubset, "Mast", "Proliferating FBs",
        outcomeCol = "disease_status", metric = "kontextual")
}

Description

Shows, for one cell type (or group), each image's proportion measured with the original labels (unweighted) and with the bootstrap confidence weights (weighted), for each level of the outcome. For the results of runProportionsPipeline() the proportion is of all cells in the image; for the results of runProportionsRelativeToParentPipeline() it is of the cells of the parent group. Each point is one image.

Usage

plotProportionBoxplot(results, cellType, spe, outcomeCol)

Arguments

Value

An interactive plot, a girafe widget that needs the ggiraph package.

Examples

data(lungSubset)
data(lungCellTypeTree)
data(lungConfList)
res = runProportionsPipeline(lungSubset, lungCellTypeTree, lungConfList,
    outcomeCol = "disease_status", subject = "subject")
rel = runProportionsRelativeToParentPipeline(lungSubset, lungCellTypeTree,
    lungConfList, outcomeCol = "disease_status", subject = "subject")
if (requireNamespace("ggiraph", quietly = TRUE)) {
    plotProportionBoxplot(res, "Arteriole", lungSubset, "disease_status")
    plotProportionBoxplot(rel, "Arteriole", lungSubset, "disease_status")
}

Description

Plots the cells of one image, highlighting a cell type or group, or showing bootstrap confidence scores.

Usage

plotWeightedSpatialPair(
  spe,
  cellTypeTree,
  confList,
  imageID,
  groupA,
  subtypeA,
  groupB,
  subtypeB,
  encoding = c("alpha", "colour"),
  cellSize = 1
)

plotHardSpatialPair(
  spe,
  imageID,
  subtypeA,
  subtypeB,
  cellTypeTree = NULL,
  cellSize = 1
)

plotWeightedSpatial(
  spe,
  cellTypeTree,
  confList,
  imageID,
  targetGroup,
  targetSubtype,
  cellSize = 1
)

plotOriginalLabelSpatial(spe, cellTypeTree, imageID, targetGroup, cellSize = 1)

Arguments

Value

A ggplot object.

Examples

data(lungSubset)
data(lungCellTypeTree)
data(lungConfList)
plotHardSpatialPair(lungSubset, "VUHD113", "Mast", "Proliferating FBs",
    lungCellTypeTree)
plotWeightedSpatial(lungSubset, lungCellTypeTree, lungConfList, "VUHD113",
    "Endothelial lineage", "Arteriole")
plotOriginalLabelSpatial(lungSubset, lungCellTypeTree, "VUHD113",
    "Endothelial lineage")
if (requireNamespace("ggnewscale", quietly = TRUE)) {
    plotWeightedSpatialPair(lungSubset, lungCellTypeTree, lungConfList,
        "VUHD113", "Granulocytes", "Mast", "Fibroblast lineage",
        "Proliferating FBs")
}

Description

Tests, for every pair of leaf cell types and groups, whether the co-localisation of the two differs between the two conditions of an outcome, once with the original labels and once with the bootstrap confidences as soft cell type weights. The co-localisation of each image is the cross-type L-function, computed by spicyR (spicyR::spicy(method = "image")), and relative co-localisation is Statial's Kontextual test (spicyR::kontextualEngine()).

Usage

runColocalisationModels(
  spe,
  cellTypeTree,
  confList,
  outcomeCol,
  r = 50,
  subject = NULL,
  cores = 1,
  parentOverrides = character(0),
  autoOverrideParents = TRUE,
  cellType = "cellType",
  cellID = "cellID",
  imageID = "imageID",
  x = "x",
  y = "y"
)

Arguments

Details

Co-localisation of a pair from -> to in an image is the cross-type L-function: how strongly the from cells are attracted to (positive values) or avoid (negative values) the to cells within radius r, with 0 meaning no attraction or repulsion. The area of an image is that of the square around its cells, and cells near the border are edge corrected. Each pair is then tested by a linear model of its per-image values on the outcome (the images are the units); if some subjects have several images, spicyR fits a mixed model with the subject as a random effect instead.

With weights (see confidenceWeights()), a cell of a bootstrapped group belongs to each subtype of the group with its bootstrap confidence: each cell contributes to a label in proportion to its weight, and the density of a label is its summed weight per unit area. With the original labels (weights of 0 and 1) this is the unweighted L-function.

Relative co-localisation compares the co-localisation of from with to to that of from with a parent population of to, so that attraction that only arises because both cell types prefer the same region is not counted. By default the parent of a to cell type is its own group; parentOverrides and autoOverrideParents give broader parents. Pairs whose to has no parent other than itself are not tested.

A label is not paired with itself, nor with a group that contains it.

Value

A list with the tests modelWeighted, modelUnweighted, modelRelativeWeighted and modelRelativeUnweighted (one row per pair: estimate, the difference for the second level of the outcome compared with the first, its se, statistic, p.value and BH-adjusted p.adj, and for relative co-localisation the parent and side, "attract" or "avoid"), the per-image values weighted, unweighted, relativeWeighted and relativeUnweighted (columns imageID, from, to, colocalisation, and parent for the relative version), and the spicyR results themselves (spicy, a list of weighted, unweighted, relativeWeighted and relativeUnweighted), for spicyR's plotting functions.

Examples

data(lungSubset)
data(lungCellTypeTree)
data(lungConfList)
res = runColocalisationModels(lungSubset, lungCellTypeTree, lungConfList,
    outcomeCol = "disease_status", subject = "subject")
head(res$modelWeighted)

Description

Computes per-image proportions of every cell type using original labels (unweighted) and bootstrap confidences (weighted), and tests each against an outcome with a linear model. runProportionsRelativeToParentPipeline() instead expresses each subtype as a proportion of its parent group.

Usage

runProportionsPipeline(
  spe,
  cellTypeTree,
  confList,
  outcomeCol,
  skipGroups = character(0),
  subject = NULL,
  cellType = "cellType",
  cellID = "cellID",
  imageID = "imageID"
)

runProportionsRelativeToParentPipeline(
  spe,
  cellTypeTree,
  confList,
  outcomeCol,
  skipGroups = character(0),
  subject = NULL,
  cellType = "cellType",
  cellID = "cellID",
  imageID = "imageID"
)

Arguments

Details

In runProportionsPipeline() the unweighted results cover every cell type and also the whole groups that were bootstrapped. Only the subtypes of those groups have confidences, so for every other cell type (cell types without a group, and the group-level proportions) the weighted proportion is the unweighted one: their labels are not uncertain. The weighted and unweighted models therefore test the same set of cell types, and their adjusted p-values are comparable.

The lineage column of the results is the group a subtype belongs to (not the broad lineage). For cell types without a group and for the group-level proportions it is the cell type or group itself.

Value

A list with proportions, weighted, unweighted (long data.frames; only proportions and not weighted/unweighted for the relative-to-parent version), and modelWeighted and modelUnweighted (per cell type model results, where estimate is the difference for the second level of the outcome compared with the first).

Examples

data(lungSubset)
data(lungCellTypeTree)
data(lungConfList)
res = runProportionsPipeline(lungSubset, lungCellTypeTree, lungConfList,
    outcomeCol = "disease_status", subject = "subject")
head(res$modelWeighted)
data(lungSubset)
data(lungCellTypeTree)
data(lungConfList)
rel = runProportionsRelativeToParentPipeline(lungSubset, lungCellTypeTree,
    lungConfList, outcomeCol = "disease_status", subject = "subject")
head(rel$modelWeighted)

Description

Computes the proportion of each image's cells in each niche and tests every niche against a two-level outcome with a Wilcoxon or t-test.

Usage

testNicheOutcome(
  spe,
  nicheUnweighted,
  nicheWeighted,
  outcomeCol,
  test = c("wilcox", "ttest"),
  subject = NULL,
  cellID = "cellID",
  imageID = "imageID"
)

Arguments

Details

The estimate is the difference in the niche proportion for the second level of the outcome compared with the first (alphabetical order for text, factor order for factors), for both tests, as in runColocalisationModels(). For test = "wilcox" this is the Hodges-Lehmann estimate of the shift in location. Each image is one observation, so every image must come from a different subject if subject is given.

Value

A list with testUnweighted, testWeighted (one row per niche) and propUnweighted, propWeighted (per image niche proportions, to pass to nicheProportionBoxplot()).

Examples

data(lungSubset)
data(lungCellTypeTree)
data(lungConfList)
niches = runNicheIdentification(lungSubset, lungCellTypeTree, lungConfList,
    r = 50, k = 6)
res = testNicheOutcome(lungSubset, niches$nicheUnweighted,
    niches$nicheWeighted, outcomeCol = "disease_status", subject = "subject")
res$testWeighted