Skip to contents

Choosing UpSet vs Venn vs Network

Source: vignettes/v04_upset_vs_venn_vs_network.Rmd
v04_upset_vs_venn_vs_network.Rmd

Choosing UpSet vs Venn vs Network

vennDiagramLab ships three complementary visualizations of the same underlying region structure. This vignette explains when to use which.

library(vennDiagramLab)
result <- analyze(load_sample("dataset_real_cancer_drivers_4"))
length(result@dataset@set_names)
#> [1] 4

Quick guidance

# of sets Recommended primary view Why
2 Venn obvious, area-proportional possible
3 Venn classic three-circle layout reads instantly
4 Venn (Edwards) still readable as a Venn
5–6 UpSet Venn becomes hard to read; UpSet bars are clearer
7+ UpSet (primary) + Network (relationships) Venn is essentially unusable

For high set counts (5+), the Network view adds something neither representation provides: it shows the pairwise relationships as a graph, where edge weight is intersection size or significance.

Venn 

svg <- render_venn_svg(result, title = "4-set Venn (cancer drivers)")
nchar(svg)
#> [1] 6473

The SVG is plain text — embed it in a notebook with htmltools::HTML(svg) or save to disk and reference from Markdown.

UpSet 

render_upset(result, sort_by = "size", color_mode = "depth")
#> Warning: Using `size` aesthetic for lines was deprecated in ggplot2 3.4.0.
#> ℹ Please use `linewidth` instead.
#> ℹ The deprecated feature was likely used in the ComplexUpset package.
#>   Please report the issue at
#>   <https://github.com/krassowski/complex-upset/issues>.
#> This warning is displayed once per session.
#> Call `lifecycle::last_lifecycle_warnings()` to see where this warning was
#> generated.

(The chunk above is gated on R >= 4.6 because the CRAN release of ComplexUpset (1.3.3) is incompatible with ggplot2 >= 4.0 on older R.)

Network 

render_network(result, edge_metric = "intersection")

Each node is a set, sized by inclusive cardinality. Each edge is a pair, weighted by the chosen edge_metric ("intersection", "jaccard", "fold_enrichment", or "overlap_coefficient"). Edges below the significance threshold are colored differently.

When the three views disagree

Sometimes a region looks “small” on a Venn but lights up bright on a Network because the fold-enrichment is high relative to expectation. That’s not a contradiction — Venn shows raw counts, Network can show normalized strength. Use both.

What’s next