tests/testthat/test-celda_C.R

In campbio/celda: CEllular Latent Dirichlet Allocation

# celda_C
library(celda)
library(SingleCellExperiment)
library(Matrix)
context("Testing celda_C")

sceceldaCSim <- simulateCells("celda_C", K = 10)
counts(sceceldaCSim) <- as(counts(sceceldaCSim), "dgCMatrix")
scesf <- selectFeatures(sceceldaCSim)
K <- S4Vectors::metadata(sceceldaCSim)$celda_simulateCellscelda_C$K
counts <- SummarizedExperiment::assay(scesf, "counts")
sce <- celda_C(scesf,
    sampleLabel = sampleLabel(scesf),
    K = K,
    algorithm = "EM",
    verbose = FALSE,
    nchains = 1)
factorized <- factorizeMatrix(sce)

# celda_C
test_that(desc = "Testing simulation and celda_C model", {
    #expect_equal(typeof(counts), "integer")
    expect_true(all(sweep(factorized$counts$sample,
        2,
        colSums(factorized$counts$sample),
        "/") == factorized$proportions$sample))
    expect_true(ncol(factorized$proportions$module) == K)
    expect_true(all(is.numeric(logLikelihoodHistory(sce))))
    expect_equal(max(logLikelihoodHistory(sce)), bestLogLikelihood(sce))
})

# clusterProbability
test_that(desc = "Testing clusterProbability with celda_C", {
    expect_true(all(round(rowSums(clusterProbability(sce)[[1]])) == 1))
})

# celdaGridSearch and perplexity calculations
test_that(desc = "Testing celdaGridSearch with celda_C", {
    celdaCResList <- celdaGridSearch(scesf,
        model = "celda_C",
        nchains = 2,
        paramsTest = list(K = c(5, 6)),
        paramsFixed = list(sampleLabel = sampleLabel(scesf)),
        maxIter = 2,
        verbose = FALSE,
        bestOnly = FALSE,
        perplexity = FALSE)
    expect_error(celdaGridSearch(scesf,
        model = "celda_C",
        paramsTest = list(K = c(4, 5), M = c(3, 4)),
        paramsFixed = list(sampleLabel = sampleLabel(scesf)),
        bestOnly = FALSE),
        paste0("The following elements in 'paramsTest' are not arguments of",
            " 'celda_C': M"))

    expect_error(celdaGridSearch(scesf,
        model = "celda_C",
        nchains = 1,
        maxIter = 1,
        paramsTest = list(K = c(4, 5), sampleLabel = "Sample"),
        paramsFixed = list(sampleLabel = sampleLabel(scesf))),
        paste0("Setting parameters such as 'z.init', 'y.init', and",
            " 'sampleLabel' in 'paramsTest' is not currently supported."))

    expect_error(celdaGridSearch(scesf,
        model = "celda_C",
        nchains = 1,
        maxIter = 1,
        paramsTest = list(),
        paramsFixed = list(sampleLabel = sampleLabel(scesf))),
        paste0("The following arguments are not in 'paramsTest' or",
            " 'paramsFixed' but are required for 'celda_C': K"))

    expect_error(celdaGridSearch(scesf,
        model = "celda_C",
        nchains = 1,
        maxIter = 1,
        paramsTest = list(K = c(9, 10)),
        paramsFixed = list(sampleLabel = sampleLabel(scesf),
            xxx = "xxx")),
        paste0("The following elements in 'paramsFixed' are not arguments",
            " of 'celda_C': xxx"))

    expect_true(class(S4Vectors::metadata(
        altExp(celdaCResList))$celda_grid_search)[1] == "celdaList")
    expect_equal(names(runParams(celdaCResList)),
        c("index", "chain", "K", "seed", "logLikelihood"))
    expect_error(plotGridSearchPerplexity(celdaCResList))

    celdaCResList <- resamplePerplexity(celdaCResList, numResample = 2)
    expect_equal(is.null(altExp(
        celdaCResList)@metadata$celda_grid_search@perplexity),
        FALSE)
    expect_true(is(celdaCResList, "SingleCellExperiment"))
    expect_error(resamplePerplexity(celdaCResList, numResample = "2"),
        "The 'numResample' parameter needs to be an integer greater than 0.")

    plotObj <- plotGridSearchPerplexity(celdaCResList)
    expect_is(plotObj, "ggplot")

    celdaCResIndex1 <- subsetCeldaList(celdaCResList, params = list(index = 1))

    expect_true(all("celda_grid_search" %in%
            names(altExp(celdaCResList)@metadata) &&
            "celda_parameters" %in% names(altExp(celdaCResIndex1)@metadata)))

    expect_error(subsetCeldaList(celdaCResList, params = list(K = 11)))
    expect_error(subsetCeldaList(celdaCResList, params = list(K = 5, M = 10)))

    celdaCResK5 <- subsetCeldaList(celdaCResList, params = list(K = 5))
    sce2 <- selectBestModel(celdaCResK5)
    res <- perplexity(sce)
    res2 <- perplexity(sce, newCounts = as.matrix(counts + 1))

    expect_error(res <- perplexity(sce, newCounts = counts[-1, ]))
})

# # logLikelihood
# test_that(desc = "Testing logLikelihood.celda_C", {
#     expect_lt(logLikelihood(sce), 0)
#     fakeZ <- celdaCSim$z
#     fakeZ[1] <- celdaCSim$K + 1
#     expect_error(logLikelihood(model = "celda_C",
#         z = fakeZ,
#         counts = celdaCSim$counts,
#         K = celdaCSim$K,
#         alpha = 1,
#         beta = 1,
#         sampleLabel = sampleLabel(sceceldaCSim)),
#         "An entry in z contains a value greater than the provided K.")
# })
#
# # Gibbs sampling
# test_that(desc = "Testing celda_C with Gibbs sampling", {
#     res <- celda_C(sceceldaCSim,
#         sampleLabel = sampleLabel(sceceldaCSim),
#         K = celdaCSim$K,
#         algorithm = "Gibbs",
#         maxIter = 5,
#         nchain = 1)
#     expect_is(res, "celda_C")
# })
#
# # normalizeCounts
# test_that(desc = paste0("Making sure normalizeCounts doesn't change",
#     " dimensions of counts matrix"), {
#         normCounts <- normalizeCounts(celdaCSim$counts)
#         expect_equal(dim(normCounts), dim(celdaCSim$counts))
#         expect_equal(rownames(normCounts), rownames(celdaCSim$counts))
#         expect_equal(colnames(normCounts), colnames(celdaCSim$counts))
#         expect_error(normalizeCounts(celdaCSim$counts,
#             transformationFun = "scale"),
#             "'transformationFun' needs to be of class 'function'")
#         expect_error(normalizeCounts(celdaCSim$counts, scaleFun = "scale"),
#             "'scaleFun' needs to be of class 'function'")
#     })
#
# # recodeClusterZ
# test_that(desc = "Testing recodeClusterZ with celda_C", {
#     expect_error(recodeClusterY(celdaMod = modelC,
#         from = c(1, 2, 3, 4, 5),
#         to = c(5, 4, 3, 2, 1)))
#     expect_error(recodeClusterZ(celdaMod = modelC,
#         from = NULL,
#         to = ""))
#     expect_error(recodeClusterZ(celdaMod = modelC,
#         from = c(1, 2, 3, 4, 5),
#         to = c(1, 2, 3, 4, 6)))
#     expect_error(recodeClusterZ(celdaMod = modelC,
#         from = c(1, 2, 3, 4, 6),
#         to = c(1, 2, 3, 4, 5)))
#     newRecoded <- recodeClusterZ(celdaMod = modelC,
#         from = c(1, 2, 3, 4, 5),
#         to = c(5, 4, 3, 2, 1))
#     expect_equal(modelC@clusters$z == 1, newRecoded@clusters$z == 5)
# })
#
# # compareCountMatrix
# # test_that(desc = "Testing CompareCountMatrix with celda_C", {
# #     expect_true(compareCountMatrix(counts = celdaCSim$counts,
# #         celdaMod = modelC))
# #
# #     lessCells <- celdaCSim$counts[, seq(100)]
# #     expect_error(compareCountMatrix(counts = lessCells, celdaMod = modelC),
# #         paste0("The provided celda object was generated from a counts",
# #         " matrix with a different number of cells than the one provided."))
# #
# #     countsMatrixError <- matrix(data = 1,
# #         nrow = nrow(celdaCSim$counts),
# #         ncol = ncol(celdaCSim$counts))
# #     expect_false(compareCountMatrix(counts = countsMatrixError,
# #         celdaMod = modelC,
# #         errorOnMismatch = FALSE))
# #     expect_error(compareCountMatrix(counts = countsMatrixError,
# #         celdaMod = modelC,
# #         errorOnMismatch = TRUE))
# # })
#
# # topRank
# test_that(desc = "Checking topRank to see if it runs without errors", {
#     topRank <- topRank(matrix = factorized$proportions$module,
#     threshold = NULL)
#     expect_equal(names(topRank), c("index", "names"))
#     topRank <- topRank(matrix = factorized$proportions$module, n = 1000)
# })
#
# # plotHeatmap
# test_that(desc = "Testing plotHeatmap with celda_C", {
#     expect_error(plotHeatmap(counts = celdaCSim$counts, z = modelC@params$K),
#         "Length of z must match number of columns in counts matrix")
#     expect_error(plotHeatmap(
#         counts = celdaCSim$counts,
#         z = modelC@clusters$z,
#         scaleRow = modelC),
#         "'scaleRow' needs to be of class 'function'")
#     expect_error(plotHeatmap(counts = celdaCSim$counts,
#         z = modelC@clusters$z,
#         trim = 3),
#         paste0("'trim' should be a 2 element vector specifying the lower and",
#             " upper boundaries"))
# })
#
# # plotHeatmap with annotationCell
# # test_that(desc = "Testing plotHeatmap with celda_C, including annotations",
# #     {
# #     annot <- as.data.frame(c(rep(x = 1,
# #         times = ncol(celdaCSim$counts) - 100),
# #         rep(x = 2, 100)))
# #
# #     rownames(annot) <- colnames(celdaCSim$counts)
# #     expect_equal(names(plotHeatmap(
# #         celdaMod = modelC,
# #         counts = celdaCSim$counts,
# #         annotationCell = annot,
# #         z = modelC@clusters$z)),
# #         c("tree_row", "tree_col", "gtable"))
# #
# #     rownames(annot) <- NULL
# #     expect_equal(names(plotHeatmap(celdaMod = modelC,
# #         counts = celdaCSim$counts,
# #         annotation.feature = as.matrix(annot),
# #         z = modelC@clusters$z)),
# #         c("tree_row", "tree_col", "gtable"))
# #
# #     rownames(annot) <- rev(colnames(celdaCSim$counts))
# #     expect_error(plotHeatmap(celdaMod = modelC,
# #         counts = celdaCSim$counts,
# #         annotationCell = annot,
# #         z = modelC@clusters$z),
# #         paste0("Row names of 'annotationCell' are different than the",
# #             " column names of 'counts'")
# #     )
# # })
#
# # celdaHeatmap
# # test_that(desc = "Testing celdaHeatmap with celda_C", {
# #     expect_equal(names(celdaHeatmap(celdaMod = modelC,
# #         counts = celdaCSim$counts)),
# #         c("tree_row", "tree_col", "gtable"))
# # })
#
# # celdaProbabilityMap
# # test_that(desc = "Testing celdaProbabiltyMap.celda_C for sample level", {
# #     plotObj <- celdaProbabilityMap(counts = celdaCSim$counts,
# #         celdaMod = modelC,
# #         level = "sample")
# #     expect_true(!is.null(plotObj))
# #
# #     ## Without a sample label
# #     modelC <- celda_C(celdaCSim$counts,
# #         sampleLabel = NULL,
# #         K = celdaCSim$K,
# #         maxIter = 5,
# #         nchain = 1)
# #     plotObj <- celdaProbabilityMap(counts = celdaCSim$counts,
# #         celdaMod = modelC,
# #         level = "sample")
# #     expect_true(!is.null(plotObj))
# # })
#
# test_that(desc = paste0("Testing celdaTsne with celda_C when model class is",
#     "changed, should error"), {
#         modelX <- modelC
#         class(modelX) <- "celda_X"
#         expect_error(celdaTsne(counts = celdaCSim$counts,
#             celdaMod = modelX,
#             maxCells = length(modelC@clusters$z),
#             minClusterSize = 10),
#             "unable to find")
#     })
#
# test_that(desc = "Testing celdaTsne with celda_C including all cells", {
#     tsne <- celdaTsne(counts = celdaCSim$counts,
#         celdaMod = modelC,
#         maxCells = length(modelC@clusters$z),
#         minClusterSize = 10)
#     plotObj <- plotDimReduceCluster(tsne[, 1],
#         tsne[, 2],
#         modelC@clusters$z,
#         labelClusters = TRUE)
#     expect_true(ncol(tsne) == 2 & nrow(tsne) == length(modelC@clusters$z))
#     expect_true(!is.null(plotObj))
# })
#
# test_that(desc = paste0("Testing celdaTsne with celda_C including",
#     " a subset of cells"), {
#     expect_success(expect_error(tsne <- celdaTsne(counts = celdaCSim$counts,
#         celdaMod = modelC,
#         maxCells = 50,
#         minClusterSize = 50)))
#     tsne <- celdaTsne(counts = celdaCSim$counts,
#         celdaMod = modelC,
#         maxCells = 100,
#         minClusterSize = 10)
#     plotObj <- plotDimReduceCluster(tsne[, 1], tsne[, 2], modelC@clusters$z)
#     expect_true(ncol(tsne) == 2 &
#             nrow(tsne) == length(modelC@clusters$z) &&
#             sum(!is.na(tsne[, 1])) == 100)
#     expect_true(!is.null(plotObj))
# })
#
# test_that(desc = paste0("Testing celdaUmap with celda_C when model class is",
#     " changed, should error"), {
#         modelX <- modelC
#         class(modelX) <- "celda_X"
#         expect_error(celdaUmap(counts = celdaCSim$counts,
#             celdaMod = modelX,
#             maxCells = length(modelC@clusters$z),
#             minClusterSize = 10),
#             "unable to find")
#     })
#
# # test_that(desc = "Testing celdaUmap with celda_C including all cells", {
# #     umap <- celdaUmap(counts = celdaCSim$counts,
# #         celdaMod = modelC,
# #         maxCells = length(modelC@clusters$z),
# #         minClusterSize = 10)
# #     plotObj <- plotDimReduceCluster(umap[, 1], umap[, 2], modelC@clusters$z)
# #     expect_true(ncol(umap) == 2 &
# #             nrow(umap) == length(modelC@clusters$z))
# #     expect_true(!is.null(plotObj))
# # })
#
# test_that(desc = paste0("Testing celdaUmap with celda_C including",
#     " a subset of cells"), {
#     expect_success(expect_error(umap <- celdaUmap(
#         counts = celdaCSim$counts,
#         celdaMod = modelC,
#         maxCells = 50,
#         minClusterSize = 50)))
#     umap <- celdaUmap(counts = celdaCSim$counts,
#         celdaMod = modelC,
#         maxCells = 100,
#         minClusterSize = 10)
#     plotObj <- plotDimReduceCluster(umap[, 1], umap[, 2], modelC@clusters$z)
#     expect_true(ncol(umap) == 2 &
#             nrow(umap) == length(modelC@clusters$z) &&
#             sum(!is.na(umap[, 1])) == 100)
#     expect_true(!is.null(plotObj))
# })
#
# # featureModuleLookup
# test_that(desc = "Testing featureModuleLookup with celda_C", {
#     expect_error(featureModuleLookup(celdaCSim$counts, modelC, "test_feat"))
# })
#
# # .cCSplitZ
# test_that(desc = "Testing error checking for .cCSplitZ", {
#     r <- simulateCells("celda_C",
#         S = 1,
#         CRange = c(50, 100),
#         K = 2)
#     dc <- .cCDecomposeCounts(r$counts, r$sampleLabel, r$z, r$K)
#     res <- .cCSplitZ(r$counts,
#         dc$mCPByS,
#         dc$nGByCP,
#         dc$nCP,
#         s = as.integer(r$sampleLabel),
#         z = r$z,
#         K = r$K,
#         nS = dc$nS,
#         nG = dc$nG,
#         alpha = 1,
#         beta = 1,
#         zProb = NULL,
#         minCell = 1000)
#     expect_true(grepl("Cluster sizes too small", res$message))
# })
#
# test_that(desc = "Testing perplexity.celda_C", {
#     expect_true(is.numeric(perplexity(celdaCSim$counts, modelC)))
#
#     class(modelC) <- c("celda_CG")
#     expect_error(perplexity.celda_C(celdaCSim$counts, modelC),
#         "could not find function \"perplexity.celda_C\"")
# })