BioVenn – an R and Python package for the comparison and visualization of biological lists using area-proportional Venn diagrams

2.1.colorfulVennPlot

The first package is ‘colorfulVennPlot’ [15]. This package can create 2-circle and 3-circle Venn diagrams, and use ellipses for diagrams of 4 sets. Only the 2-circle diagrams can be made area-proportional, but the user needs to calculate the circles’ sizes and overlap by using the separate ‘resizeCircles’ function.

2.2.eulerr

A second package is ‘eulerr’ [14], which can generate area-proportional Euler diagrams. A Euler diagram is a generalization of a Venn diagram, relaxing the criterion that all interactions need to be represented. In practice, both terms are used interchangeably. This package uses both ellipses and circles.

2.3.nVennR

A third package is ‘nVennR’ [16]. This package can create “quasi-proportional Venn and Euler diagrams” for an unlimited number of sets. For a large number of sets, the algorithm might be very slow, because it needs to run many simulation cycles. Because of the resulting complicated shapes, the diagrams might not be easy to read.

2.4.venn

A fourth package is ‘venn’ [6], which can generate Venn diagrams up to 7 sets, but not in an area-proportional manner. For more than three sets, it uses pre-set polygon shapes.

2.5.VennDiagram

The most popular package at this moment is ‘VennDiagram’ [3]. This package can generate Venn and Euler diagrams of up to five sets, but these are not area-proportional, unless the user calculates the radii and distances between the circles by him- or herself, and passes these numbers through to one of the draw.*.venn functions.

2.6.venneuler

A sixth package is ‘venneuler’ [19]. This package can create area-proportional Venn diagrams as well, if the sizes of the overlaps are passed to its venneuler function. The returned object also gives some mathematical information such as the residuals (percentage difference between input intersection area and fitted inter-section area) and stress values.

2.7.vennplot

The seventh and final package is ‘vennplot’ [20]. It can create area-proportional Venn diagrams in 2D or 3D, with two or three circles or balls. The 3D functionality is interesting (the diagram can be rotated), but the mathematics behind it is actually the same as for the 2D plot.

3.1.matplotlib-venn

The most popular package at this moment is ‘matplotlib-venn’ [18]. Its ‘venn2’ and ‘venn3’ functions can create area-proportional Venn diagrams of two and three circles, respectively. However, they don’t offer the ID mapping functionality of BioVenn, and the ‘drag-and-drop’ functionality for repositioning of titles and labels in the SVG mode of BioVenn is missing as well.

3.2.PyVenn

A second package is ‘PyVenn’ [8]. This package offers plotting of Venn diagrams of two to six circles, but these are not area-proportional like in BioVenn or Matplotlib-Venn: the shapes are always the same.

5.1.R example

The following very simple R code creates the example plot of Fig. 1, and returns the data of Table 1:

5.2.Python example

The Python code works in a very similar manner:

Note that the code in both R and Python could be even compressed into one line, by adding the lists directly to the draw.venn or draw_venn command. For improved readability we use a four-line code.

5.3.Biological ID mapping

To enable the mapping of Affymetrix or Entrez identifiers to Ensembl identifiers, the parameter ‘map2ens’ should be set to ‘TRUE’, e.g. in R:

This code creates the example plot of Fig. 2 and returns the data of Table 2. Note that, in comparison to Fig. 1 and Table 1, some lists contain more identifiers. This is because some Affymetrix IDs map to multiple Ensembl Gene IDs (possibly homologous genes). With the ‘map2ens’ function, BioVenn automatically converts the Affymetrix IDs to their corresponding Ensembl Gene IDs, and draws the Venn diagrams using the Ensembl Gene IDs. This is useful for researchers that want to do a gene-based comparison from expression data.

5.4.Overall comparison

To compare the different R and Python packages, we created Venn diagrams of a dataset showing orthologous genes that are present in human (Homo sapiens), mouse (Mus musculus) or the African clawed frog (Xenopus laevis) (available at the OMA Browser [1] through https://omabrowser.org/All/oma-groups.txt.gz). Since human and mouse are more closely related than human and Xenopus (and mouse and Xenopus), we expect that the circles of human and mouse have a larger overlap. Furthermore, Xenopus has more genes, so its circle should be larger than the circles of human and mouse.

Figure 3 shows the Venn Diagrams created in each of the R packages, in alphabetical order. For each of the plots, the colours red, green and blue were used, titles were removed, and numbers were printed in the diagram (if that option was available). The code used to generate the plots can be viewed at https://www.biovenn.nl/r_python/. We can see that the packages that create area-proportional diagrams (a, c, d, g, h) give a better impression of what the data looks like: the human and mouse circles indeed have a larger overlap than with the Xenopus circle, and the Xenopus circle is larger than the other ones. The nVennR diagram (d) might be visually less appealing, but it displays the information correctly as well. The non-area-proportional diagrams (b, e, f) need some careful reading of the numbers in the figure before they can be interpreted.

Fig. 3.

Venn diagrams created by each of the R packages: a) BioVenn, b) colorfulVennPlot, c) eulerr, d) nVennR, e) venn, f) VennDiagram, g) venneuler and h) vennplot.

Fig. 4.

Venn diagrams created by each of the Python packages: a) BioVenn, b) matplotlib-venn and c) PyVenn.

Figure 4 shows the Venn Diagrams created in each of the Python packages, in alphabetical order, with the same method as described above. Again, the area-proportional diagrams (a, b) can be understood much more easily than the non-area-proportional diagram (c).

Table 3 shows a comparison of all features of BioVenn and the seven other packages mentioned above. BioVenn is the only package that is available in both R and Python (as well as a web interface). There are packages that can generate Venn diagrams from more than three sets, but these are either not area-proportional or inaccurate. For three sets, it is sometimes impossible to create a completely accurate area-proportional Venn diagram; when more sets are added, this becomes an even larger issue. Only BioVenn has built-in biological ID mapping functionality, which earns it the prefix ‘bio’. Some programs support not only the input of IDs, but also the numbers of the sets and their overlaps. In BioVenn, these are automatically calculated from the ID lists. This also makes sure that the user cannot input mathematically impossible numbers (e.g. overlaps larger than the sets themselves). BioVenn supports a large number of output formats as well. It needs to be noted here that this table only lists built-in functionality; some functionality such as plotting to certain file formats might be possible by using other R or Python functions (e.g. the ‘matplotlib.pyplot’ functions in Python). BioVenn is the only package that supports drag-and-drop of the titles and labels (in SVG mode), which can be a very useful functionality when a set or overlap is very small compared to the rest of the figure, or when the circle title (e.g. ‘Set X’, ‘Set Y’, ‘Set Z’) overlaps with a number. BioVenn uses only circles, whereas other packages also use ellipses, polygons or even 3D balls. There are four packages (BioVenn, eulerr, PyVenn and VennDiagram) that are able to print absolute numbers or percentages in the diagram. Finally, BioVenn offers the most flexibility in formatting: title, subtitle and circle texts can be changed (as well as their fonts and colors), and the background color and the circle colors can be set.