Hypergeometric and Fisher's exact tests evaluate whether the proportion of module genes annotated with term $T$ exceeds the proportion in the eligible reference pool. If you test leaf drought tissue, thousands of root-specific, flower-specific, and embryonic genes are unexpressed and were filtered out during preprocessing.

Including those inactive genes in the background artificially suppresses the baseline frequency of leaf-expressed terms, causing widespread false-positive inflation of generic metabolic terms.

library(clusterProfiler)
library(enrichplot)

# Step 1: Define eligible network genes (All non-grey module genes)
eligible_genes <- names(module_labels)[module_colors[names(module_labels)] != "grey"]

# Step 2: Query OrgDb and retain only genes with valid Biological Process (BP) GO terms
mapped_ids <- AnnotationDbi::select(
    org.At.tair.db,
    keys = unique(eligible_genes),
    keytype = "TAIR",
    columns = c("GO", "ONTOLOGY")
)
mapped_ids <- mapped_ids[!is.na(mapped_ids$GO) & mapped_ids$ONTOLOGY == "BP", ]

# The True Statistical Background
background <- unique(mapped_ids$TAIR)
cat("Verified Non-Grey GO BP Background size:", length(background), "genes\n")

# Step 3: Intersect module genes with background
module_genes <- intersect(target_genes, background)
if (length(module_genes) == 0) {
    stop("No genes from the target module map to the GO BP background.")
}

# Step 4: Perform Over-Representation Analysis (ORA)
ego_module <- enrichGO(
    gene          = module_genes,
    universe      = background,       # Strictly controlled network universe
    OrgDb         = org.At.tair.db,
    keyType       = "TAIR",
    ont           = "BP",
    pAdjustMethod = "BH",
    pvalueCutoff  = 0.05,
    qvalueCutoff  = 0.05,
    minGSSize     = 10,
    maxGSSize     = 500,
    readable      = TRUE
)

ego_df <- as.data.frame(ego_module)
cat("\nEnriched Biological Process Terms (Top 10):\n")
print(head(ego_df[, c("ID", "Description", "GeneRatio", "BgRatio", "p.adjust")], 10))

if (nrow(ego_df) > 0) {
    # Collapse redundant terms with Wang semantic similarity cutoff
    ego_simple <- simplify(ego_module, cutoff = 0.7, by = "p.adjust", select_fun = min)
    
    # Generate Publication Dotplot
    p_dot <- dotplot(ego_simple, showCategory = 15) +
        ggtitle(paste("GO Biological Process Enrichment:", target_ME, "(Drought Response)")) +
        theme_minimal(base_size = 12) +
        theme(
            plot.title = element_text(face = "bold", size = 14),
            axis.text.y = element_text(color = "black")
        )
    print(p_dot)
} else {
    message("No GO BP terms achieved statistical significance at BH q < 0.05.")
}