plot_gsea.leadingedge.HCL: GSEA Leading Edge HCL plot.

In drmjc/metaGSEA: Meta-analysis of GSEA analyses, including intra- and inter-experiment comparisons

Description

This function compares genesets in terms of the overlap of the leading edge genes within each geneset. This information is presented as a hierarchical clustering plot, where distance is measured proportional to the number of genes in common (see jacquard). The user can either pass in a leading edge list (ie a list where each element is a vector of gene symbols), or (as will be more common), an entire leading edge list (eg 1434 elements for c2_all), in conjunction with a GSEA top table (tt), coupled with a threshold, such as the top 50, or those with FDR < 0.25 via the N, FDR, P, or FWER arguments.

Usage

plot_gsea.leadingedge.HCL(x, N = NULL, NES = NULL, FDR = NULL, P = NULL,
  FWER = NULL, direction = c("either", "up", "down"),
  main = "Leading edge similarities", xlab = "", yaxis = c("similarity",
  "distance"), max.distance = 1, rename.genesets = FALSE,
  label.clusters = FALSE, cluster.threshold = 0.99,
  hclust.method = "complete", ...)

Arguments

x	a GSEA object. see import.gsea
N	if not NULL, then choose the top N most significant genesets
NES	if not NULL, then choose those genesets with |NES| > the threshold
FDR	if not NULL, then choose those genesets with FDR < the threshold
P	if not NULL, then choose those genesets with nominal P < the threshold
FWER	if not NULL, then choose those genesets with FWER < the threshold
direction	include just the genesets that are “up”, “down”, or “either” means top 50 genesets that are either up or down
main	additional arguments passed to plot.
xlab	additional arguments passed to plot.
yaxis	either “similarity” or “distance”. Two closely related genesets have eg 90% similarity, and a distance of 0.1.
max.distance	(experimental) attempt to filter out genesets that are dissimilar to all other genesets. This is a distance threshold, not a similarity threshold, so values of 0.9 are a good place to start. See note.
rename.genesets	should the genesets be renamed to include the rank/fdr/direction
label.clusters	if TRUE, then draw red boxes around the clusters. see cluster.threshold.
cluster.threshold	the distance threshold if label.clusters=TRUE. 0.99 is a good default in practice.
hclust.method	the hierarchical clustering method. see hclust
...	additional arguments passed to plot.

Value

invisibly returns the hclust object.

Note

max.distance
The intuitive outcome of setting max.distance (eg 0.1) is that no genesets with a distance of >0.1 (ie a similarity of <90%) should remain in the HCL. This is not quite the case. We filter out genesets that have distance to all other genesets > 0.9 in the 'distance' object (see gsea.leadingedge.distance) which is slightly different to filtering out nodes in the resulting HCL (which is made upon the distance object). Net effect:
- You can get singletons in the filtered HCL because all of the things that the geneset was connected to have been filtered out.
- You can get genesets in a tree of significant genesets, which is itself below the distance threshold due to a similar explanation to above.

Author(s)

Mark Cowley, 2009-04-06

See Also

plot_gsea.leadingedge.HCL plclust.gsea gsea.leadingedge.distance

drmjc/metaGSEA documentation built on Aug. 8, 2020, 1:53 p.m.