# Tests for integration related fxns set.seed(42) pbmc_small <- suppressWarnings(UpdateSeuratObject(pbmc_small)) # Setup test objects ref <- pbmc_small ref <- FindVariableFeatures(object = ref, verbose = FALSE, nfeatures = 100) query <- CreateSeuratObject( counts = as.sparse( GetAssayData( object = pbmc_small[['RNA']], layer = "counts") + rpois(n = ncol(pbmc_small), lambda = 1 ) ) ) query2 <- CreateSeuratObject( counts = as.sparse( LayerData( object = pbmc_small[['RNA']], layer = "counts")[, 1:40] + rpois(n = ncol(pbmc_small), lambda = 1 ) ) ) query.list <- list(query, query2) query.list <- lapply(X = query.list, FUN = NormalizeData, verbose = FALSE) query.list <- lapply(X = query.list, FUN = FindVariableFeatures, verbose = FALSE, nfeatures = 100) query.list <- lapply(X = query.list, FUN = ScaleData, verbose = FALSE) query.list <- suppressWarnings(lapply(X = query.list, FUN = RunPCA, verbose = FALSE, npcs = 20)) anchors2 <- suppressMessages(suppressWarnings(FindIntegrationAnchors(object.list = c(ref, query.list[[1]]), k.filter = NA, verbose = FALSE))) anchors3 <- suppressMessages(suppressWarnings(FindIntegrationAnchors(object.list = c(ref, query.list), k.filter = NA, verbose = FALSE))) # Tests for IntegrateEmbeddings # ------------------------------------------------------------------------------ # context("IntegrateEmbeddings") # test_that("IntegrateEmbeddings validates properly", { # expect_error(IntegrateEmbeddings(anchorset = anchors2)) # expect_error(IntegrateEmbeddings(anchorset = anchors2, reduction = "pca", k.weight = 100)) # expect_error(IntegrateEmbeddings(anchorset = anchors2, reduction = c("pca", "pca2"), k.weight = 40)) # expect_error(IntegrateEmbeddings(anchorset = anchors2, reduction = "pca", k.weight = 40, weight.reduction = c(ref[['pca']]))) # pca3 <- RenameCells(object = ref[['pca']], new.names = paste0(Cells(ref), "_test")) # expect_error(IntegrateEmbeddings(anchorset = anchors2, reduction = "pca", k.weight = 40, # weight.reduction = c(pca3, ref[['pca']]))) # }) # # test_that("IntegrateEmbeddings with two objects default works", { # skip_on_cran() # int2 <- IntegrateEmbeddings(anchorset = anchors2, reduction = "pca", k.weight = 40, verbose = FALSE) # expect_equal(Reductions(int2), "integrated_pca") # expect_equal(sum(Embeddings(int2[['integrated_pca']])[1,]), -3.13050872287, tolerance = 1e-6) # expect_equal(sum(Embeddings(int2[['integrated_pca']])[,1]), -5.78790844887, tolerance = 1e-6) # }) # # test_that("IntegrateEmbeddings with three objects default works", { # skip_on_cran() # int3 <- IntegrateEmbeddings(anchorset = anchors3, reduction = "pca", k.weight = 40, verbose = FALSE) # expect_equal(Reductions(int3), "integrated_pca") # expect_equal(sum(Embeddings(int3[['integrated_pca']])[1,]), 0.221867815987, tolerance = 1e-6) # expect_equal(sum(Embeddings(int3[['integrated_pca']])[,1]), -16.7881409595, tolerance = 1e-6) # }) # # test_that("IntegrateEmbeddings works with specified reference objects", { # skip_on_cran() # anchors4 <- suppressMessages(suppressWarnings(FindIntegrationAnchors(object.list = c(ref, query.list), k.filter = NA, verbose = FALSE, reference = 1))) # int4 <- IntegrateEmbeddings(anchorset = anchors4, reduction = "pca", k.weight = 40, verbose = FALSE) # expect_equal(Reductions(int4), "integrated_pca") # expect_equal(sum(Embeddings(int4[['integrated_pca']])[1,]), -3.13050872287, tolerance = 1e-6) # expect_equal(sum(Embeddings(int4[['integrated_pca']])[,1]), 13.1180105492, tolerance = 1e-6) # }) # Tests for IntegrateData # ------------------------------------------------------------------------------ context("IntegrateData") test_that("IntegrateData with two objects default work", { expect_error(IntegrateData(anchorset = anchors2)) int2 <- suppressWarnings(IntegrateData(anchorset = anchors2, k.weight = 50, verbose = FALSE)) expect_true(all(Assays(int2) %in% c("integrated", "RNA"))) expect_equal(Tool(int2), "Integration") expect_equal(dim(int2[["integrated"]]), c(133, 160)) expect_equal(length(VariableFeatures(int2)), 133) expect_equal(GetAssayData(int2[["integrated"]], layer = "counts"), new("dgCMatrix")) expect_equal(GetAssayData(int2[['integrated']], layer = "scale.data"), matrix()) expect_equal(sum(GetAssayData(int2[["integrated"]], layer = "data")[1, ]), 44.97355, tolerance = 1e-3) expect_equal(sum(GetAssayData(int2[["integrated"]], layer = "data")[, 1]), 78.8965706046, tolerance = 1e-6) expect_equal(Tool(object = int2, slot = "Integration")@sample.tree, matrix(c(-1, -2), nrow = 1)) }) test_that("IntegrateData with three objects default work", { expect_error(IntegrateData(anchorset = anchors3, k.weight = 50)) int3 <- suppressWarnings(IntegrateData(anchorset = anchors3, k.weight = 25, verbose = FALSE)) expect_true(all(Assays(int3) %in% c("integrated", "RNA"))) expect_equal(Tool(int3), "Integration") expect_equal(dim(int3[["integrated"]]), c(169, 200)) expect_equal(length(VariableFeatures(int3)), 169) expect_equal(GetAssayData(int3[["integrated"]], layer = "counts"), new("dgCMatrix")) expect_equal(GetAssayData(int3[['integrated']], layer = "scale.data"), matrix()) expect_equal(sum(GetAssayData(int3[["integrated"]], layer = "data")[1, ]), 372.829, tolerance = 1e-6) expect_equal(sum(GetAssayData(int3[["integrated"]], layer = "data")[, 1]), 482.5009, tolerance = 1e-6) expect_equal(Tool(object = int3, slot = "Integration")@sample.tree, matrix(c(-2, -3, 1, -1), nrow = 2, byrow = TRUE)) }) test_that("Input validates correctly ", { expect_error(anchorset = anchors2, k.weight = 50, features.to.integrate = "BAD") expect_error(IntegrateData(anchorset = anchors2, k.weight = 50, normalization.method = "BAD")) expect_error(IntegrateData(anchorset = anchors2, k.weight = 50, weight.reduction = "BAD")) expect_error(IntegrateData(anchorset = anchors2, reductions.to.integrate = "pca")) skip_on_cran() #expect_warning(IntegrateData(anchorset = anchors2, k.weight = 50, features = c(rownames(ref), "BAD"))) #expect_warning(IntegrateData(anchorset = anchors2, k.weight = 50, dims = 1:1000)) })