Skip to contents

Extract Differentially Expressed Genes per Subtype

Source: R/extract_top_genes.R
extract_top_genes.Rd

For each molecular subtype, performs gene-level differential expression analysis comparing samples of that subtype against all other samples (one-vs-rest). Returns genes passing user-specified fold-change and adjusted p-value thresholds, sorted by significance.

Usage

extract_top_genes(
  expr,
  subtype,
  idType = "SYMBOL",
  method = c("wilcox", "t.test"),
  log2fc_cutoff = 1,
  p_cutoff = 0.05,
  min_samples = 3
)

Arguments

expr

A numeric gene expression matrix with genes in rows and samples in columns. Row names must be gene symbols (or identifiers matching idType). Values should be log2-transformed (e.g. log2(TPM + 1)).

subtype

A named character vector or factor giving the predicted subtype for each sample, as returned by classifyHNSC(outputType = "class").

idType

Character string specifying the gene identifier type. One of "SYMBOL" (default), "ENSEMBL", "ENTREZID", or "REFSEQ". Non-symbol IDs are auto-converted to gene symbols via org.Hs.eg.db.

method

Character string specifying the statistical test. One of "wilcox" (default) for Wilcoxon rank-sum test or "t.test" for Welch's t-test.

log2fc_cutoff

Numeric. Minimum absolute log2 fold-change to retain a gene (default: 1, i.e. 2-fold change).

p_cutoff

Numeric. Adjusted p-value (FDR) threshold (default: 0.05).

min_samples

Numeric. Minimum number of samples required in each group (subtype vs rest) to run the comparison (default: 3).

Value

A named list of data frames, one per subtype. Each data frame contains genes passing the thresholds, with columns:

  • gene: gene symbol

  • log2FC: log2 fold-change (positive = up in this subtype)

  • mean_expr: mean expression in this subtype

  • mean_other: mean expression in all other subtypes

  • p_value: raw p-value from the chosen test

  • adj_p_value: Benjamini-Hochberg adjusted p-value (FDR)

Details

For each subtype, the function compares in-subtype samples against all remaining samples on every gene. P-values are corrected across all genes within each subtype using the Benjamini-Hochberg procedure.

method = "wilcox" uses wilcox.test which makes no distributional assumptions. method = "t.test" uses Welch's t-test via t.test which assumes approximate normality.

Examples

if (FALSE) { # \dontrun{
data(TCGA_LUSC)
subtypes <- classifyHNSC(TCGA_LUSC, outputType = "class")
degs <- extract_top_genes(TCGA_LUSC, subtypes,
                           log2fc_cutoff = 1, p_cutoff = 0.05)
head(degs$Basal)
} # }