cnet_heatmap_guide.md

In jmw86069/multienrichjam: Analysis and Visualization of Multiple Gene Set Enrichments

Brief steps to produce a Cnet-Heatmap plot

1. Run multiEnrichMap()

2. Apply appropriate p_cutoff, and topEnrichN to reduce the number of pathways per enrichment result, topEnrichN from 10 to 20 for two enrichment results; topEnrichN 4 to 10 for four or more enrichment results.

3. Run mem_plot_folio()

cairo_pdf(file="folio_mem_DM_AffySoma_top20_23apr2020.pdf",
   height=12, width=12, pointsize=11, onefile=TRUE);
mem_dm_plots <- mem_plot_folio(mem_dm_affysoma,
   do_which=2:5,
   pathway_column_split=4,
   main="Dermatomyositis Affy-SomaLogic");
dev.off();

• Make sure do_which includes c(2,4), which produces cnet_collapsed_set

• Review the gene-protein heatmap to ensure the clusters are representative of the underlying content of the data. Adjust clustering parameters as needed, for example:

  • column_method="correlation" changes column clustering to use correlation, instead of euclidean distance, which can emphasize the top annotation enrichment colors, slightly more than the gene-protein matrix.
  • min_gene_ct=1, min_sample_ct=1 filters pathways and genes to require more than one gene per pathway, or more than one pathway per gene. Sometimes this filtering helps the clustering focus on genes/pathways which are shared rather than random singlet genes.
  • pathway_column_split=4 to adjust the number of pathway clusters, can be helpful if there are more or fewer clusters visible in the gene-pathway heatmap than are being produced by the automatic process. For large number of pathways, sometimes more clusters are necessary.
  • p_floor especially when p_cutoff is lower than 0.01, the p_floor should be set appropriately to represent the range of enrichment P-values in the top annotation color bar. Note that it may not be useful to show P-values of 1e-100 -- anything less than 1e-10 should in theory be equivalent.
  • Note the location of Cnet clusters, c("A", "B", "C", "D") because they need to be defined manually in the Cnet-Heatmap.

• (Omit this step -- now these steps are performed inside plot_cnet_heatmaps())

Extract cnet_collapsed_set as the Cnet plot to use for the Cnet-Heatmap.

ht_opt("legend_border"="black");
cnet_collapsed_set <- mem_dm_plots$cnet_collapsed_set;

# Optionally rotate igraph layout coordinates to help position labels
#cnet_collapsed_set <- rotate_igraph_layout(cnet_collapsed_set, degree=180);
#cnet_collapsed_set <- spread_igraph_labels(cnet_collapsed_set);

E(cnet_collapsed_set)$color <- "#BBBBBB99";
V(cnet_collapsed_set)$frame.color <- "#77777799";
isset <- V(cnet_collapsed_set)$nodeType %in% "Set";
V(cnet_collapsed_set)$label <- ifelse(isset,
   V(cnet_collapsed_set)$name,
   V(cnet_collapsed_set)$label);
V(cnet_collapsed_set)$label.family <- "Arial";
V(cnet_collapsed_set)$label.color <- ifelse(isset,
   "grey85",
   "grey10");
## Manually add distance for label from node
V(cnet_collapsed_set)$label.dist <- 0.8;
V(cnet_collapsed_set)$label.dist[igrep("^PR", V(cnet_collapsed_set)$name)] <- 2;

• Optionally adjust the network layout using some helpful functions:

  • rotate_igraph_layout() - rotate the layout so the clusters can be aligned with heatmaps on the outer edges.
  • spread_igraph_labels() - spreads out igraph labels based upon the angle of incoming edges, it does pretty well at reducing overlapping labels. Use nodeSortBy to change color-sorting from "left-to-right" with nodeSortBy="x", to "top-to-bottom" with nodeSortBy="-y". Use label_min_dist to increase the distance of labels from the node, where 0.5 is roughly half the node diameter, which is of course 1x radius. Use force_relayout and repulse to force a new layout with different repulse adjustment, higher repulse makes nodes more tightly grouped.

• Prepare centered expression data, whose rownames are genes that are present in the gene-pathway data above.

• Typically data should be log2-transformed, and row-centered, which can be done using jamma::centerGeneData().

  • Center each group by its associated control group, using centerGroups and controlSamples arguments. It is not recommended to scale the data, but that is available with scale="row".
  • Optionally omit control groups from the heatmaps to maximize plot space. You may want to create a plot with control groups to verify that the control groups have low variability, then re-create the plot without control groups for the final figure.

• Define isamples using colnames() of the expression matrix.

• Define igroups to be a vector of group labels, named by colnames() of the expression matrix.

• Define colorV whose names are unique values from igroups.

  • These colorV values are shown at the top of each heatmap as annotations to represent the groups in igroups. The names are displayed below the heatmap, should they should be small enough to fit in a limited space. Use legend_labels below to use a different label for the color legend.
  • Ideally these colors should also match the colors from colorV produced by multiEnrichMap(), but it is not required.
  • Note that when the color is matched between the sample group and the gene node, a box is drawn around the heatmap panel to indicate these genes were used in the enrichment test.

• Run plot_cnet_heatmap().

{cairo_pdf(file="DM_Lupus_GeneProtein_MultiEnrich_Heatmap_27apr2020.pdf",
   height=9, width=10, pointsize=12);
all_samples_dmsle2
igroups_dlgp <- gsubOrdered("[a-z_ ]+", "", all_cohort_type[all_samples_dmsle2])
colorV_dlgp <- nameVector(mem_dmlupus_affysoma$colorV,
   levels(factor(igroups_dlgp[all_samples_dmsle2])));
pch_dlgp2 <- plot_cnet_heatmaps(mem_folio=mem_dmlupus_affysoma_plots6,
   iexprs_ctr=all_pergene_exprs_ctr,
   isamples=all_samples_dmsle2,
   rotate_degrees=15,
   layout_reflect="y",
   igroups=igroups_dlgp,
   set_names=c("A","B","D","C"),
   set_panel_row=list(c(1:5), c(1:4), c(6:8), c(6:8)),
   cnet_mar=c(0.5, 6.2, 0.7, 6),
   vjust=c(0.9, 0.9, -0.5, -0.5),
   strwrap_width=c(60, 33, 36, 36),
   colorV=colorV_dlgp)
dev.off();}

• There are numerous arguments to this function, intended to help configure the many details of the figure.
• iexprs_ctr - the expression data matrix, intended to show centered data, but in fact any numeric data is accepted.
• allowed_degree=c(1,2) uses any gene node with 1 or 2 connections for each cluster heatmap. When allowed_degree=1 the heatmap will include genes that connect only to that pathway cluster, and no other pathway clusters. Sometimes it is useful to focus on specific genes, to make sure there are no changes visible in the heatmap across the other groups. If there are changes across other groups, the specificity may be an artifact of choosing too stringent a statistical threshold for gene hits prior to running the enrichment test.
• legend_labels are used to provide a custom label for the colors, helpful when colorV has abbreviated labels.
• cnet_mar - adjust margins c(bottom, left, top, right) around the central Cnet plot, to reduce overlaps across labels.
• vjust - vertical justification of pathway labels for each cluster, where 1 positions labels at the top of the top layout cell, 0 positions labels at the top of the next layout cell lower, and -1 positions labels at the top of the layout cell two cells lower.
• strwrap_width is used to define the word wrap width for each pathway cluster. Sometimes it helps to make some sections wider or narrower than others, to reduce label overlaps.
• set_names vector of pathway clusters to use, in order. These names should reflect the cluster names such as c("A", "B", "C", "D"), but should be in order they appear in the Cnet layout.
• Define set_panel_row and set_panel_col as the range of rows and columns for each heatmap. They should be positioned relatively close to the Cnet pathway clusters, with height relative to the number of genes.

• use_shadowText=TRUE uses a slight outline effect around node labels, which makes it easier to read over light/dark background colors.

• Save the same data to Excel:

## Convert to data.frame format
dlgp_df_cnet <- get_shared_pathways(ipa_pathways_all,
   groups=mem_dmlupus_affysoma$enrichLabels,
   cnet=mem_dmlupus_affysoma_plots6_v2$cnet_collapsed_set,
   subset_cnet=TRUE);
## save to xlsx
save_shared_pathways(file="DM_Lupus_GeneProtein_MultiEnrich_Heatmap_top4_20apr2020.xlsx",
   df=dlgp_df_cnet,
   sheetName="multienrich_DM_Lupus_GP",
   colorSub=colorSub);

jmw86069/multienrichjam documentation built on Feb. 7, 2024, 12:58 a.m.