r‑statistics.co by Selva Prabhakaran

About

UpSet Plot in R: Visualize Set Intersections Beyond Venn Diagrams

An UpSet plot replaces Venn diagrams for data with more than 3 sets — it shows every intersection as a bar chart, with a matrix below marking which sets belong to each intersection. In R, the UpSetR package creates them with a single upset() call.

Introduction

Venn diagrams work beautifully for 2-3 sets. Beyond that, the overlapping circles become unreadable — try drawing a 6-set Venn diagram and you'll see why researchers needed an alternative.

UpSet plots solve this systematically. Instead of overlapping circles, they use:

  1. A bar chart (top) — each bar is an intersection, sized by how many elements belong to it.
  2. A matrix (bottom) — dots and lines show which sets are included in each intersection.
  3. Set size bars (left) — total size of each individual set.

The result is a precise, readable display of every intersection between any number of sets — 4, 8, even 20 sets become manageable.

Common use cases: gene ontology overlaps in bioinformatics, survey respondents who chose multiple answers, product features selected by customers, or any analysis where elements can belong to multiple groups simultaneously.

How do you read an UpSet plot?

Before building one, let's understand the layout. An UpSet plot has three panels:

  • Top bar chart: Each bar = one intersection. The tallest bar = most common intersection.
  • Matrix grid: Each column = one intersection. Filled dots show which sets are in that intersection. A single dot = elements only in that set. Connected dots = elements in all connected sets simultaneously.
  • Left bar chart: Total elements in each set (regardless of intersections).

For example, if you have movies that can be Drama, Comedy, or Action:

  • A bar with only the Drama dot filled = movies that are only Drama (not Comedy, not Action).
  • A bar with Drama and Comedy dots connected = movies that are both Drama AND Comedy.

How do you create a basic UpSet plot in R?

UpSetR uses upset() which takes a data frame in binary format — one column per set, values 0/1.

library(UpSetR) # Built-in movies dataset from UpSetR # Each row is a movie, columns are genres (0/1 membership) data(movies) # Basic UpSet plot upset( movies, sets = c("Action", "Comedy", "Drama", "Romance", "Thriller"), mb.ratio = c(0.55, 0.45) # ratio of bar chart to matrix panel height )


  


  






movies is a built-in dataset in UpSetR — 3,000+ movies with binary genre columns. sets specifies which sets to include and their order (bottom-to-top in the matrix). mb.ratio controls the vertical split between the bar chart and matrix.



Try it: Remove the sets argument to include all columns automatically. Then change mb.ratio = c(0.7, 0.3) to give more space to the bar chart.



How do you sort intersections by size?



By default, intersections are sorted by degree (number of sets in the intersection). Sorting by frequency (size) puts the most common intersections first — usually more useful.





  


    
# Sort by frequency (most common intersections first)
upset(
  movies,
  sets        = c("Action", "Comedy", "Drama", "Romance", "Thriller"),
  order.by    = "freq",           # sort by intersection size
  decreasing  = TRUE,             # largest first
  nsets       = 5,                # include 5 sets
  nintersects = 15,               # show top 15 intersections
  mb.ratio    = c(0.55, 0.45)
)


    


      

      

    


    


  


  






order.by = "freq" sorts bars by size. nintersects = 15 limits to the 15 largest intersections — helpful when many possible intersections exist but most are tiny.



Try it: Change order.by = "freq" to order.by = "degree" (default) — the bars are now sorted by how many sets each intersection spans. Is this more or less useful than frequency sorting for this dataset?



How do you customize colors in an UpSet plot?



UpSetR has limited ggplot2-style styling, but it does support bar color customization with main.bar.color, sets.bar.color, and matrix.color.





  


    
# Custom bar colors
upset(
  movies,
  sets        = c("Action", "Comedy", "Drama", "Romance", "Thriller"),
  order.by    = "freq",
  nintersects = 12,
  mb.ratio    = c(0.55, 0.45),
  main.bar.color = "#1565C0",    # intersection bar color
  sets.bar.color = "#E53935",    # set size bar color (left panel)
  matrix.color   = "#1565C0",    # dot color in matrix
  shade.color    = "#EEF2FF",    # shade alternating rows
  text.scale     = c(1.5, 1.3, 1, 1, 1.5, 1.2)  # text scaling per panel
)


    


      

      

    


    


  


  






text.scale controls font size for 6 elements in this order: intersection size labels, intersection axis label, set size labels, set axis labels, set names, numbers above dots.



Try it: Change main.bar.color = "#2E7D32" (green) and sets.bar.color = "#F57C00" (orange). Also try removing shade.color to remove the alternating row shading.



How do you highlight a specific intersection?



query lets you highlight specific intersections — useful for calling out the most important pattern in a presentation.





  


    
# Highlight the Drama-only intersection
upset(
  movies,
  sets     = c("Action", "Comedy", "Drama", "Romance", "Thriller"),
  order.by = "freq",
  nintersects = 12,
  mb.ratio    = c(0.55, 0.45),
  queries = list(
    list(
      query  = intersects,
      params = list("Drama"),      # only Drama (no other sets)
      color  = "#E53935",          # highlight color
      active = TRUE,               # fill the bar
      query.name = "Drama Only"
    )
  )
)


    


      

      

    


    


  


  






intersects is a built-in query function in UpSetR. params = list("Drama") selects the intersection where only Drama is active. active = TRUE fills the bar; active = FALSE draws an outline only.



Try it: Change params = list("Drama") to params = list("Action", "Drama") to highlight the Action + Drama co-occurrence intersection.



How do you create an UpSet plot from a binary matrix?



If your data isn't already in the UpSetR built-in format, convert it from a list of sets or a binary membership matrix.





  


    
# Create binary data from scratch: survey responses
# Each row = one respondent, each column = a feature they use
set.seed(42)
n <- 200

bin_df <- data.frame(
  Email     = sample(0:1, n, replace = TRUE, prob = c(0.2, 0.8)),
  Dashboard = sample(0:1, n, replace = TRUE, prob = c(0.3, 0.7)),
  Reports   = sample(0:1, n, replace = TRUE, prob = c(0.4, 0.6)),
  API       = sample(0:1, n, replace = TRUE, prob = c(0.7, 0.3)),
  Mobile    = sample(0:1, n, replace = TRUE, prob = c(0.5, 0.5))
)

upset(
  bin_df,
  sets        = c("Email", "Dashboard", "Reports", "API", "Mobile"),
  order.by    = "freq",
  nintersects = 15,
  mb.ratio    = c(0.6, 0.4),
  main.bar.color = "#1565C0",
  sets.bar.color = "#43A047"
)


    


      

      

    


    


  


  






The data frame just needs columns with 0/1 values — one row per element, one column per set.



Try it: Add keep.order = TRUE inside upset() to keep the set order as specified in sets instead of automatically sorting by set size.



Common Mistakes and How to Fix Them



Mistake 1: Data not in binary format



upset() requires 0/1 binary columns. Lists of members need conversion first.





  


    
# Convert list of sets to binary matrix
set_list <- list(
  SetA = c("gene1", "gene2", "gene3"),
  SetB = c("gene2", "gene4", "gene5"),
  SetC = c("gene1", "gene5", "gene6")
)

all_elements <- unique(unlist(set_list))
bin_mat <- data.frame(
  lapply(set_list, function(s) as.integer(all_elements %in% s)),
  row.names = all_elements
)


    


      

      

    


    


  


  






Mistake 2: Including too many intersections



With 6+ sets, there are 2^6 = 64 possible intersections. Use nintersects to show only the most common ones.





  


    
upset(df, nintersects = 20, order.by = "freq")


    


      

      

    


    


  


  






Mistake 3: Confusing "set size" with "intersection size"



The left bar chart shows total set membership (including overlap). The top bar chart shows exclusive intersection size. These are different: the Drama bar on the left is all drama movies; the Drama bar on top is only movies that are exclusively Drama.



Mistake 4: Using Venn diagrams for > 3 sets



With 4+ sets, Venn diagrams become overlapping circles that are nearly impossible to read. Switch to UpSet plots for any dataset with 4+ groups.



Mistake 5: Not labeling axes clearly



UpSetR's defaults have minimal labeling. Add ylab = "Intersection Size" and use text.scale to increase font sizes for readability.





  


    
upset(df, ylab = "Intersection Size", text.scale = c(1.5, 1.3, 1.2, 1, 1.5, 1))


    


      

      

    


    


  


  






Practice Exercises



Exercise 1: Build your own binary matrix



Create a data frame of 100 customers who purchased different product categories (Electronics, Books, Clothing, Food, Sports — each with random 0/1 membership). Create an UpSet plot sorted by frequency showing the top 10 intersections.




Show solution



  

    
library(UpSetR)

set.seed(99)
n <- 100
df <- data.frame(
  Electronics = rbinom(n, 1, 0.4),
  Books       = rbinom(n, 1, 0.5),
  Clothing    = rbinom(n, 1, 0.35),
  Food        = rbinom(n, 1, 0.6),
  Sports      = rbinom(n, 1, 0.25)
)

upset(
  df,
  sets        = c("Electronics", "Books", "Clothing", "Food", "Sports"),
  order.by    = "freq",
  nintersects = 10,
  mb.ratio    = c(0.6, 0.4),
  main.bar.color = "#1565C0"
)

    

      
      
    

    

  

  








Exercise 2: Using the built-in movies dataset



Using data(movies) from UpSetR, create an UpSet plot for these 6 genres: Action, Comedy, Drama, Romance, Documentary, Animation. Highlight the Comedy-only intersection in red.




Show solution



  

    
library(UpSetR)
data(movies)

upset(
  movies,
  sets        = c("Action", "Comedy", "Drama", "Romance", "Documentary", "Animation"),
  order.by    = "freq",
  nintersects = 15,
  mb.ratio    = c(0.55, 0.45),
  queries     = list(
    list(
      query      = intersects,
      params     = list("Comedy"),
      color      = "#E53935",
      active     = TRUE,
      query.name = "Comedy Only"
    )
  )
)

    

      
      
    

    

  

  








Summary







Argument
Effect






sets
Which sets to include (columns in data frame)




order.by = "freq"
Sort intersections by size (most common first)




order.by = "degree"
Sort by number of sets in each intersection




nintersects
Limit to top N intersections




nsets
Limit to top N sets




mb.ratio
Panel height ratio (bar chart vs matrix)




main.bar.color
Intersection bar color




sets.bar.color
Set size bar color (left panel)




queries
Highlight specific intersections




text.scale
Font size control for each panel







When to use UpSet plots:



    


  • 4+ sets where Venn diagrams are unreadable
    • 


  • You need exact counts for every intersection
    • 


  • The most common intersections are the key finding
    • 





When to use Venn diagrams:



    


  • 2-3 sets only
    • 


  • Your audience is unfamiliar with UpSet plots
    • 


  • Proportional area matters more than exact counts (Euler diagrams)
    • 





FAQ



What is the difference between UpSetR and ComplexUpset? UpSetR is the original package — simple, fast, single function call. ComplexUpset rebuilds UpSet plots in ggplot2, giving full ggplot2 theme and layer control but with more complex syntax. Start with UpSetR; move to ComplexUpset if you need ggplot2 integration.



How do I convert a list of set members to binary format? Use fromList() from UpSetR: upset(fromList(list(SetA = c("a","b"), SetB = c("b","c"))), ...). This converts a list of members into the binary matrix format automatically.



Can I make interactive UpSet plots in R? Yes — the upsetjs package creates interactive UpSet plots for R Markdown and Shiny. The ComplexUpset package also supports some interactivity via plotly.



Why does my UpSet plot not show some intersections? By default nintersects = 40. If you have few observations, some intersections may have 0 members and are hidden. Increase nintersects or check your data for empty categories.



How do I export an UpSet plot to a file? Wrap in pdf() / dev.off() or use png(): png("upset.png", width=10, height=6, units="in", res=150); upset(df, ...); dev.off().



References



    


  • UpSetR CRAN package: github.com/hms-dbmi/UpSetR
    • 


  • Lex A., et al. (2014). UpSet: Visualization of Intersecting Sets. IEEE TVCG.
    • 


  • Conway J.R., et al. (2017). UpSetR: an R package for the visualization of intersecting sets. Bioinformatics.
    • 


  • upset.app — Official UpSet visualization website
    • 





What's Next?



    


  • R Waffle Chart — visualize counts as grids of unit squares
    • 


  • Heatmap in R — encode a matrix of values as a color grid
    • 


  • R Correlation Matrix Plot — visualize pairwise correlations across many variables
    • 




        


        
      


      

        

        
© 2016- Selva Prabhakaran.
          This work is licensed under the Creative Commons License.