hdWGCNA

hdWGCNA is an R package designed for performing weighted gene co-expression network analysis (WGCNA) on high-dimensional transcriptomics data, specifically single-cell RNA-seq or spatial transcriptomics.

Key features include:

• Modular Design: Can construct context-specific co-expression networks across cellular and spatial hierarchies.
• Module Identification: Identifies modules of highly co-expressed genes and provides biological context via statistical testing and knowledge sources.
• Transcription Factor Analysis: Supports Transcription Factor Regulatory Network Analysis.
• Data Compatibility: Uses datasets formatted as Seurat objects.

For practical implementation, users should refer to the hdWGCNA in single-cell data tutorial or the hdWGCNA in spatial transcriptomics data tutorial.

The visual structure of a network depends heavily on the layout algorithm. ggraph supports several layouts, including:

• stress (default)
• kk (Kamada Kawai)
• layout_with_fr() (Fruchterman-Reingold)
• layout_as_tree()
• layout_nicely()
• layout_in_circle()

You can also provide a custom layout using 2D coordinates (e.g., from a UMAP) by using ggraph::create_layout(graph, custom_layout_df).

# 1. default layout
p1 <- ggraph(graph) + 
  geom_edge_link(aes(alpha=weight, color=edge_color)) + 
  geom_node_point(aes(color=module)) + 
  scale_colour_manual(values=mod_cp) + 
  scale_edge_colour_manual(values=mod_cp) + 
  ggtitle("layout = 'stress' (auto)") + 
  NoLegend()

# 2. Kamada Kawai (kk) layout
graph2 <- graph; E(graph)$weight <- E(graph)$weight + 0.0001
p2 <- ggraph(graph, layout='kk', maxiter=100) + 
  geom_edge_link(aes(alpha=weight, color=edge_color)) + 
  geom_node_point(aes(color=module)) + 
  scale_colour_manual(values=mod_cp) + 
  scale_edge_colour_manual(values=mod_cp) + 
  ggtitle("layout = 'kk'") + 
  NoLegend()

# 3. igraph layout_with_fr
p3 <- ggraph(graph, layout=layout_with_fr(graph)) + 
  geom_edge_link(aes(alpha=weight, color=edge_color)) + 
  geom_node_point(aes(color=module)) + 
  scale_colour_manual(values=mod_cp) + 
  scale_edge_colour_manual(values=mod_cp) + 
  ggtitle("layout_with_fr()") + 
  NoLegend()

# 4. igraph layout_as_tree
p4 <- ggraph(graph, layout=layout_as_tree(graph)) + 
  geom_edge_link(aes(alpha=weight, color=edge_color)) + 
  geom_node_point(aes(color=module)) + 
  scale_colour_manual(values=mod_cp) + 
  scale_edge_colour_manual(values=mod_cp) + 
  ggtitle("layout_as_tree()") + 
  NoLegend()

# 5. igraph layout_nicely
p5 <- ggraph(graph, layout=layout_nicely(graph)) + 
  geom_edge_link(aes(alpha=weight, color=edge_color)) + 
  geom_node_point(aes(color=module)) + 
  scale_colour_manual(values=mod_cp) + 
  scale_edge_colour_manual(values=mod_cp) + 
  ggtitle("layout_nicely()") + 
  NoLegend()

# 6. igraph layout_in_circle
p6 <- ggraph(graph, layout=layout_in_circle(graph)) + 
  geom_edge_link(aes(alpha=weight, color=edge_color)) + 
  geom_node_point(aes(color=module)) + 
  scale_colour_manual(values=mod_cp) + 
  scale_edge_colour_manual(values=mod_cp) + 
  ggtitle("layout_in_circle()") + 
  NoLegend()

Consensus co-expression network analysis differs from the standard workflow by constructing individual networks across distinct datasets (e.g., different biological sexes or species) and then computing an integrated co-expression network. This allows for the identification of conserved networks across conditions or the construction of a unified network from multiple datasets.

The workflow involves:

1. Metacell Construction: Using MetacellsByGroups with the relevant metadata columns in group.by to ensure groups are partitioned correctly.
2. Expression Setup: Using SetMultiExpr instead of SetDatExpr to create separate expression matrices for each group.
3. Soft Power Selection: Using TestSoftPowersConsensus to find optimal thresholds for each individual matrix.
4. Network Construction: Using ConstructNetwork with consensus=TRUE to build individual networks and integrate them.

# Example of the consensus workflow logic
seurat_obj <- SetMultiExpr(seurat_obj, group_name = "ASC", group.by = "cell_type", multi.group.by ="msex")
seurat_obj <- TestSoftPowersConsensus(seurat_obj)
seurat_obj <- ConstructNetwork(seurat_obj, soft_power=c(8,5), consensus=TRUE, tom_name = "Sex_Consensus")

To address the sparsity of gene expression in Visium ST datasets, hdWGCNA uses a data aggregation approach to produce spatial metaspots. Unlike single-cell metacells which aggregate based on transcriptomes, metaspots aggregate neighboring spots based on spatial coordinates.

Use MetaspotsByGroups to perform this aggregation. The group.by parameter allows you to slice the Seurat object to construct metaspots separately for different groups (e.g., different ST slides, clusters, or anatomical regions).

Workflow:

1. Initialize hdWGCNA setup with SetupForWGCNA.
2. Aggregate spots using MetaspotsByGroups.
3. Normalize the resulting metaspots with NormalizeMetacells.
4. (Optional) Extract the metaspot object using GetMetacellseurat_vhd.

seurat_obj <- SetupForWGCNA(
  seurat_obj,
  gene_select = "fraction",
  fraction = 0.05,
  wgcna_name = "vis"
)

seurat_obj <- MetaspotsByGroups(
  seurat_obj,
  group.by = c("region"),
  ident.group = "region",
  assay = 'Spatial'
)
seurat_obj  <- NormalizeMetacells(seurat_obj)

# To extract the metaspot object:
m_obj <- GetMetacellseurat_vhd(seurat_obj)