testDA_edgeR: Test for differential abundance: method 'diffcyt-DA-edgeR'

In lmweber/diffcyt: Differential discovery in high-dimensional cytometry via high-resolution clustering

Test for differential abundance: method 'diffcyt-DA-edgeR'

Description

Calculate tests for differential abundance of cell populations using method 'diffcyt-DA-edgeR'

Usage

testDA_edgeR(
  d_counts,
  design,
  contrast,
  trend_method = "none",
  min_cells = 3,
  min_samples = NULL,
  normalize = FALSE,
  norm_factors = "TMM"
)

Arguments

d_counts	SummarizedExperiment object containing cluster cell counts, from calcCounts.
design	Design matrix, created with createDesignMatrix. See createDesignMatrix for details.
contrast	Contrast matrix, created with createContrast. See createContrast for details.
trend_method	Method for estimating dispersion trend; passed to function estimateDisp from edgeR package. Default = "none". (See estimateDisp help file from edgeR package for other options.)
min_cells	Filtering parameter. Default = 3. Clusters are kept for differential testing if they have at least min_cells cells in at least min_samples samples.
min_samples	Filtering parameter. Default = number of samples / 2, which is appropriate for two-group comparisons (of equal size). Clusters are kept for differential testing if they have at least min_cells cells in at least min_samples samples.
normalize	Whether to include optional normalization factors to adjust for composition effects (see details). Default = FALSE.
norm_factors	Normalization factors to use, if normalize = TRUE. Default = "TMM", in which case normalization factors are calculated automatically using the 'trimmed mean of M-values' (TMM) method from the edgeR package. Alternatively, a vector of values can be provided (the values should multiply to 1).

Details

Calculates tests for differential abundance of clusters, using functions from the edgeR package.

This method uses the edgeR package (Robinson et al. 2010, Bioinformatics; McCarthy et al. 2012, Nucleic Acids Research) to fit models and calculate moderated tests at the cluster level. Moderated tests improve statistical power by sharing information on variability (i.e. variance across samples for a single cluster) between clusters. By default, we use the option trend.method = "none" to calculate dispersions, since the dispersion-mean relationship typically does not resemble RNA-sequencing data; see edgeR User's Guide. The statistical methods implemented in the edgeR package were originally designed for the analysis of gene expression data such as RNA-sequencing counts. Here, we apply these methods to cluster cell counts.

The experimental design must be specified using a design matrix, which can be created with createDesignMatrix. Flexible experimental designs are possible, including blocking (e.g. paired designs), batch effects, and continuous covariates. See createDesignMatrix for more details.

The contrast matrix specifying the contrast of interest can be created with createContrast. See createContrast for more details.

Filtering: Clusters are kept for differential testing if they have at least min_cells cells in at least min_samples samples. This removes clusters with very low cell counts across conditions, to improve power.

Normalization for the total number of cells per sample (library sizes) and total number of cells per cluster is automatically performed by the edgeR functions. Optional normalization factors can also be included to adjust for composition effects in the cluster cell counts per sample. For example, in an extreme case, if several additional clusters are present in only one condition, while all other clusters are approximately equally abundant between conditions, then simply normalizing by the total number of cells per sample will create a false positive differential abundance signal for the non-differential clusters. (For a detailed explanation in the context of RNA sequencing gene expression, see Robinson and Oshlack, 2010.) Normalization factors can be calculated automatically using the 'trimmed mean of M-values' (TMM) method (Robinson and Oshlack, 2010), implemented in the edgeR package (see also the edgeR User's Guide for details). Alternatively, a vector of values can be provided (the values should multiply to 1).

Value

Returns a new SummarizedExperiment object, with differential test results stored in the rowData slot. Results include raw p-values (p_val) and adjusted p-values (p_adj) from the edgeR moderated tests, which can be used to rank clusters by evidence for differential abundance. Additional output columns from the edgeR tests are also included. The results can be accessed with the rowData accessor function.

Examples

# For a complete workflow example demonstrating each step in the 'diffcyt' pipeline, 
# see the package vignette.

# Function to create random data (one sample)
d_random <- function(n = 20000, mean = 0, sd = 1, ncol = 20, cofactor = 5) {
  d <- sinh(matrix(rnorm(n, mean, sd), ncol = ncol)) * cofactor
  colnames(d) <- paste0("marker", sprintf("%02d", 1:ncol))
  d
}

# Create random data (without differential signal)
set.seed(123)
d_input <- list(
  sample1 = d_random(), 
  sample2 = d_random(), 
  sample3 = d_random(), 
  sample4 = d_random()
)

# Add differential abundance (DA) signal
ix_DA <- 801:900
ix_cols_type <- 1:10
d_input[[3]][ix_DA, ix_cols_type] <- d_random(n = 1000, mean = 2, ncol = 10)
d_input[[4]][ix_DA, ix_cols_type] <- d_random(n = 1000, mean = 2, ncol = 10)

experiment_info <- data.frame(
  sample_id = factor(paste0("sample", 1:4)), 
  group_id = factor(c("group1", "group1", "group2", "group2")), 
  stringsAsFactors = FALSE
)

marker_info <- data.frame(
  channel_name = paste0("channel", sprintf("%03d", 1:20)), 
  marker_name = paste0("marker", sprintf("%02d", 1:20)), 
  marker_class = factor(c(rep("type", 10), rep("state", 10)), 
                        levels = c("type", "state", "none")), 
  stringsAsFactors = FALSE
)

# Prepare data
d_se <- prepareData(d_input, experiment_info, marker_info)

# Transform data
d_se <- transformData(d_se)

# Generate clusters
d_se <- generateClusters(d_se)

# Calculate counts
d_counts <- calcCounts(d_se)

# Create design matrix
design <- createDesignMatrix(experiment_info, cols_design = "group_id")

# Create contrast matrix
contrast <- createContrast(c(0, 1))

# Test for differential abundance (DA) of clusters
res_DA <- testDA_edgeR(d_counts, design, contrast)

lmweber/diffcyt documentation built on March 19, 2024, 5:24 a.m.