1. Dense Matrix Scaling Laws & The RAM Exhaustion Danger

A frequent disaster in WGCNA occurs when analysts attempt to export the complete Topological Overlap Matrix (TOM) directly to disk for Cytoscape. Because the TOM is a completely dense, double-precision symmetric matrix, memory consumption scales quadratically with node count:

Moreover, rendering $> 1,000$ nodes in Cytoscape creates a visual "hairball" where topological clarity is lost. The standard protocol is to extract a top-hub subnetwork (e.g., $N = 300$) restricted to the module of interest.

2. Safe Intramodular Subnetwork Extraction & Export Workflow

Calculate the exact memory requirement in R, extract the top-ranked hub genes by $|k_{ME}|$, recalculate the intramodular TOM exclusively on those genes, and export weighted edge and node files:

MAX_NODES <- 300L

# hub_df contains target_genes ranked by absolute kME
selected_genes <- head(hub_df$gene, MAX_NODES)
selected_genes <- intersect(selected_genes, target_genes)

stopifnot(
    length(selected_genes) > 1,
    all(selected_genes %in% colnames(expr_mat))
)

# Approximate double precision storage formula
n_nodes <- length(selected_genes)
estimated_gib <- (n_nodes^2 * 8) / (1024^3)
cat("Subnetwork Node Count:", n_nodes, "\n")
cat("Memory for single dense matrix:", round(estimated_gib * 1024, 2), "MiB\n")

# Subset expression matrix to selected hub genes
expr_mod <- expr_mat[, selected_genes, drop = FALSE]

# Calculate intramodular adjacency and TOM
adj_mod <- adjacency(expr_mod, power = soft_power, type = "signed")
TOM_mod <- TOMsimilarity(adj_mod, TOMType = "signed")
dimnames(TOM_mod) <- list(selected_genes, selected_genes)

# Prepare Node Annotations
node_annot <- hub_df[hub_df$gene %in% selected_genes, ]
rownames(node_annot) <- node_annot$gene
node_annot <- node_annot[selected_genes, ]

# Export to Cytoscape (Threshold >= 0.10 filters weak edges)
edge_file <- paste0("Cytoscape_edges_", target_color, ".txt")
node_file <- paste0("Cytoscape_nodes_", target_color, ".txt")

exportNetworkToCytoscape(
    TOM_mod,
    edgeFile = edge_file,
    nodeFile = node_file,
    weighted = TRUE,
    threshold = 0.10,
    nodeNames = selected_genes,
    nodeAttr = node_annot$symbol
)

cat("✓ Network files exported successfully:\n")
cat("  - Edge Table:", edge_file, "\n")
cat("  - Node Table:", node_file, "\n")

3. Cytoscape Visual Styling Protocol

Import the exported network files into Cytoscape ($3.10+$) and apply standard publication aesthetics:

1. File Import: Navigate to File → Import → Network from File... and select the edge table. Set column 1 as Source Node, column 2 as Target Node, and column 3 (weight) as Edge Attribute.
2. Node Table Merge: Navigate to File → Import → Table from File... and import the node file, keying on node name.
3. Layout Selection: Apply Layout → Prefuse Force Directed Layout → weight or yFiles Organic Layout.
4. Node Size & Labels: Under the Style control panel, map Node Size continuously to $|k_{ME}|$ ($30 \to 80\text{ px}$), making central hub genes visually prominent. Set Node Label to Gene Symbol (or TAIR ID).
5. Node Color: Map Node Fill Color to $\log_2\text{Fold Change}$ (from DESeq2 differential analysis) using a diverging blue-white-red gradient.
6. Edge Transparency: Map Edge Transparency continuously to TOM weight, ensuring weak peripheral ties fade into the background.