Sc dge deseq2

single_cell/dge/deseq2_function.R

@@ -0,0 +1,91 @@
+library(edgeR)
+library(DESeq2)
+library(tidyverse)
+### DGE analysis using DESeq2
+#
+#' Run DGE between 2 groups using DESeq2.
+#' @param counts A feature (row) x sample (column) raw count matrix.  Only samples intended for DGE analysis should be included in \code{counts}.
+#' @param metadata Sample metadata, with rownames containing column names of \code{counts}.
+#' @param dge_by Name of \code{metadata} column to use for DGE comparion. Must have exactly two levels.
+#' @param case_group Group to be used as numerator in log2FC calculation.
+#' @param reference_group Group to be used as denominator in log2FC calculation.
+#' @param fixed_effects A vector of \code{metadata} column names to use as fixed effects.
+#' @param return_model A boolean indicating whether the function should return the model (TRUE) or the data.frame of differential expression analysis results.
+run_deseq2 <- function(counts, metadata, dge_by, case_group, reference_group, fixed_effects=NULL, return_model = FALSE) {
+
+  # Commong input checks
+  contrasts = .input_checks(counts, metadata, dge_by, case_group, reference_group, fixed_effects)
+
+  # Method specific input checks
+  # None for DESeq2
+
+  counts = .remove_low_expression(counts=counts, metadata=metadata, dge_by=dge_by, dge_groups=c(case_group, reference_group), min_frac=0.6)
+
+  # Make formula string
+  formula_str = paste("~", dge_by)
+  if(! is.null(fixed_effects)) {
+    for(fe in fixed_effects) {
+      formula_str = paste(formula_str, "+", fe)
+    }
+  }
+
+  # Perform DGE analysis
+  dds = DESeqDataSetFromMatrix(counts, colData = metadata, design = as.formula( formula_str ))
+  dds = DESeq(dds)
+
+  # Extract and format DGE results
+  res = results(dds, contrast = c(dge_by, case_group, reference_group)) %>% as.data.frame()
+  res = arrange(res, padj)
+
+  res = dplyr::rename(res, "log2FC"="log2FoldChange", "aveExpr"="baseMean", "pval"="pvalue", "padj"="padj")
+
+  if(return_model) {
+    return(dds)
+  } else {
+    return(res)
+  }
+}
+
+
+
+
+
+
+
+
+### DGE analysis using DESeq2 across cell-types
+#
+#' Run DGE between 2 groups using DESeq2 across cell-types.
+#' @param counts A feature (row) x sample (column) raw count matrix.  Only samples intended for DGE analysis should be included in \code{counts}.
+#' @param metadata Sample metadata, with rownames containing column names of \code{counts}.
+#' @param dge_by Name of \code{metadata} column to use for DGE comparion. Must have exactly two levels.
+#' @param case_group Group to be used as numerator in log2FC calculation.
+#' @param reference_group Group to be used as denominator in log2FC calculation.
+#' @param fixed_effects A vector of \code{metadata} column names to use as fixed effects.
+#' @param return_model A boolean indicating whether the function should return the model (TRUE) or the data.frame of differential expression analysis results.
+run_deseq2_within_cells <- function(counts, metadata, cell_by, cell_targets, dge_by, case_group, reference_group, fixed_effects=NULL, return_model = FALSE) {
+
+  # Check inputs
+  cell_targets = .input_checks_within_cells(metadata, cell_by, cell_target, dge_by, case_group, reference_group)
+
+  dge_res = list()
+  for(cell in cell_targets) {
+    # Assuming the rownames of metadata are in counts columns
+    cell_counts = cell_counts[ , rownames(metadata)[ metadata[,cell_by] == cell ] ]
+    cell_metadata = metadata[ metadata[,cell_by] == cell, ]
+    dge_res[[cell]] = run_deseq2(counts=cell_counts, 
+				 metadata=cell_metadata, 
+				 dge_by=dge_by, 
+				 case_group=case_group, 
+				 reference_group=reference_group, 
+				 fixed_effects=fixed_effects, 
+				 return_model = return_model)
+    if(! return_model)
+      dge_res[[cell]][, cell_by] = cell
+  }
+
+  if(! return_model)
+  dge_res <- bind_rows(dge_res)
+  return(dge_res)
+}
+