tests/testthat/test-downsampling.R

In keyes-timothy/tidytof: Analyze High-dimensional Cytometry Data Using Tidy Data Principles

library(dplyr)
library(purrr)
library(tidytof)

data(phenograph_data)

# tof_downsample_constant ------------------------------------------------------

### downsampling without grouping

test_that("Output shape is correct.", {
    result <-
        phenograph_data |>
        tof_downsample_constant(num_cells = 100L)

    # number of columns is unchanged
    expect_identical(ncol(result), ncol(phenograph_data))

    # number of rows is correct
    expect_identical(nrow(result), 100L)
})

### downsampling with grouping

test_that("Output shape is correct.", {
    result <-
        phenograph_data |>
        tof_downsample_constant(group_cols = sample_name, num_cells = 100L)

    result_2 <-
        phenograph_data |>
        tof_downsample_constant(
            group_cols = c(sample_name, phenograph_cluster),
            num_cells = 100L
        )

    # number of columns is unchanged
    expect_identical(ncol(result), ncol(phenograph_data))
    expect_identical(ncol(result_2), ncol(phenograph_data))


    # number of rows is correct
    expect_identical(nrow(result), 300L)
    expect_identical(nrow(result), 300L)
})


# tof_downsample_prop ----------------------------------------------------------

### downsampling without grouping

test_that("Output shape is correct.", {
    result <-
        phenograph_data |>
        tof_downsample_prop(prop_cells = 0.1)

    # number of columns is unchanged
    expect_identical(ncol(result), ncol(phenograph_data))

    # number of rows is correct
    expect_equal(nrow(result), nrow(phenograph_data) * 0.1)
})

### downsampling with grouping

test_that("Output shape is correct.", {
    result <-
        phenograph_data |>
        tof_downsample_prop(group_cols = sample_name, prop_cells = 0.1)

    result_2 <-
        phenograph_data |>
        tof_downsample_prop(
            group_cols = c(sample_name, phenograph_cluster),
            prop_cells = 0.1
        )

    # number of columns is unchanged
    expect_identical(ncol(result), ncol(phenograph_data))
    expect_identical(ncol(result_2), ncol(phenograph_data))

    # number of rows is correct
    expect_equal(nrow(result), 0.1 * nrow(phenograph_data))
    expect_equal(nrow(result), 0.1 * nrow(phenograph_data))
})


# tof_downsample_density ----------------------------------------------------------

### downsampling without grouping

###### KNN density estimation

test_that("Output shape is correct using KNN density estimation.", {
    result <-
        phenograph_data |>
        tof_downsample_density(target_percentile = 0.05, density_cols = c(cd19, cd11b))

    result_2 <-
        phenograph_data |>
        tof_downsample_density(target_num_cells = 100L)

    result_3 <-
        phenograph_data |>
        tof_downsample_density(target_prop_cells = 0.1, density_estimation_method = "sum_distance")

    # number of columns is unchanged
    expect_identical(ncol(result), ncol(phenograph_data))
    expect_identical(ncol(result_2), ncol(phenograph_data))
    expect_identical(ncol(result_3), ncol(phenograph_data))

    # number of rows is smaller after density-dependent downsampling
    expect_true(nrow(result) < nrow(phenograph_data))
    expect_true(nrow(result_2) < nrow(phenograph_data))
    expect_true(nrow(result_3) < nrow(phenograph_data))
})

###### spade density estimation

test_that("Output shape is correct using KNN density estimation.", {
    result <-
        phenograph_data |>
        tof_downsample_density(target_percentile = 0.05, density_cols = c(cd19, cd11b))

    result_2 <-
        phenograph_data |>
        tof_downsample_density(target_num_cells = 100L)

    result_3 <-
        phenograph_data |>
        tof_downsample_density(target_prop_cells = 0.1, density_estimation_method = "sum_distance")

    # number of columns is unchanged
    expect_identical(ncol(result), ncol(phenograph_data))
    expect_identical(ncol(result_2), ncol(phenograph_data))
    expect_identical(ncol(result_3), ncol(phenograph_data))

    # number of rows is smaller after density-dependent downsampling
    expect_true(nrow(result) < nrow(phenograph_data))
    expect_true(nrow(result_2) < nrow(phenograph_data))
    expect_true(nrow(result_3) < nrow(phenograph_data))
})

### downsampling with grouping

###### KNN density estimation

test_that("Output shape is correct using KNN density estimation.", {
    result <-
        phenograph_data |>
        tof_downsample_density(
            group_cols = sample_name,
            target_percentile = 0.05,
            density_cols = c(cd19, cd11b)
        )

    result_2 <-
        phenograph_data |>
        tof_downsample_density(
            group_cols = c(sample_name, phenograph_cluster),
            target_prop_cells = 0.1,
            # density_cols = where,
            density_estimation_method = "mean_distance"
        )

    result_3 <-
        phenograph_data |>
        tof_downsample_density(
            group_cols = c(sample_name, phenograph_cluster),
            target_num_cells = 100L,
            density_estimation_method = "sum_distance"
        )

    # number of columns is unchanged
    expect_identical(ncol(result), ncol(phenograph_data))
    expect_identical(ncol(result_2), ncol(phenograph_data))
    expect_identical(ncol(result_3), ncol(phenograph_data))

    # number of rows is correct
    expect_true(nrow(result) < nrow(phenograph_data))
    expect_true(nrow(result_2) < nrow(phenograph_data))
    expect_true(nrow(result_3) < nrow(phenograph_data))

    expect_true(nrow(result) > 0)
    expect_true(nrow(result_2) > 0)
    expect_true(nrow(result_3) > 0)
})

###### spade density estimation

test_that("Output shape is correct with SPADE density estimation.", {
    result <-
        phenograph_data |>
        tof_downsample_density(
            group_cols = sample_name,
            density_cols = c(cd19, cd11b),
            target_percentile = 0.05,
            density_estimation_method = "spade"
        )

    result_2 <-
        phenograph_data |>
        tof_downsample_density(
            group_cols = c(sample_name, phenograph_cluster),
            target_prop_cells = 0.1,
            density_estimation_method = "spade"
        )

    result_3 <-
        phenograph_data |>
        tof_downsample_density(
            group_cols = c(sample_name, phenograph_cluster),
            target_num_cells = 100L,
            density_estimation_method = "spade"
        )

    # number of columns is unchanged
    expect_identical(ncol(result), ncol(phenograph_data))
    expect_identical(ncol(result_2), ncol(phenograph_data))
    expect_identical(ncol(result_3), ncol(phenograph_data))

    # number of rows is correct
    expect_true(nrow(result) < nrow(phenograph_data))
    expect_true(nrow(result_2) < nrow(phenograph_data))
    expect_true(nrow(result_3) < nrow(phenograph_data))

    # all results have more than 0 rows
    expect_true(nrow(result) > 0)
    expect_true(nrow(result_2) > 0)
    expect_true(nrow(result_3) > 0)

    # expect error when arguments for a different density estimation method
    # are provided
    expect_error(
        phenograph_data |>
            tof_downsample_density(
                group_cols = sample_name,
                target_prop_cells = 0.1,
                num_neighbors = 10,
                density_estimation_method = "spade"
            )
    )
})