VennDiagramWeb: a web application for the generation of highly customizable Venn and Euler diagrams

Data visualization is a growing and important area of computational biology that demands high quality images which highlight the critical aspects of data. To elucidate all essential features of the data, one must perform a wide range of adjustments to various aspects of the plot, which can be a time-consuming process. Having fine-grained control over the parameters which define colours, fonts, label placements, element sizes, overall resolution, etc. leads to more effective plots which can convey necessary details in a publication-ready manner.

Pipelines facilitate automated, robust and reproducible data generation and analysis. Plotting is an important tool for both validation and reporting of results. Incorporating effective plots into these pipelines requires code that has been written specifically for each plot, as there is no single approach which can be applied to varied datasets. As a result, bioinformaticians often engage in long cycles of sequentially modifying plotting code, executing it, and observing the ultimate figure, until an optimum is reached. This process is inefficient and time-consuming.

Venn and Euler diagrams are used frequently in computational biology to visualize the interactions between multiple sets of data. In genomics especially, a common assay is to compare gene lists occurring from separate analyses [1], such as contrasting lists of differentially abundant RNAs following drug treatments or lists of mutated genes across disease types. Venn diagrams are typically depicted as partially intersecting circles or other closed curves such that there are 2ⁿ separated regions as depicted by overlapping closed curves [2]. While Venn diagrams always depict all 2ⁿ possible regions, Euler diagrams can omit regions under which there are zero values in that region’s subset. This allows Euler diagrams to be less visually complex, by depicting only a subset of all possible regions [3]. VennDiagramWeb facilitates the creation of both Venn and Euler diagrams, using the argument euler.d and scaled (both default TRUE). By default, VennDiagramWeb will create an Euler diagram where possible, displaying only regions containing one or more values. Users can force Venn diagrams only by setting both euler.d = FALSE and scaled = FALSE (Fig. 1).

The R statistical programming language has widespread use in the bioinformatics field, and so we developed VennDiagram to generate plots in this language [4]. The initial release has proven to be robust and useful, and has garnered 186 citations. As of June 8, 2016 the package has been downloaded from the Comprehensive R Archive Network (CRAN) over 75,000 times since its release in March 2011. Over half of these (>40,000) occurred in 2015 alone, highlighting growing popularity [5].

We believe a graphic user interface for VennDiagram could bring the package to a wider audience and enhance workflows for pipeline developers by providing a real-time framework for plotting optimization. There are many existing web interfaces for creating Venn diagrams, including Venny [6], BioVenn [7], GeneVenn [8], and those from the CRP-Sante Microarray Centre [9] and the Universiteit Gent [10]. These tools perform the necessities of creating a Venn diagram, but are missing many features required to create completely customized publication-quality plots, and have no means of exporting code for integration in large scale analysis pipelines.

Our first step was to improve upon the existing VennDiagram R package [4]. A series of changes were made to enhance code quality, including significant refactoring and documentation and exposure of several helper functions. Major feature additions included the ability to create quintuple Venn diagrams. These are highly complex figures, but maintain symmetry and are still easily interpretable (Fig. 2). A parameter to allow users to set a scale by which the areas and labels of the categories will be adjusted to was added. The ability to display proportions of the total population contained within the areas as percentages was also introduced. Many users requested a feature to display a text table of the partitions of the Venn diagram, which is now supported by the package. Users can also now specify an argument which will force the Venn diagram to only consider unique elements in each category when tabulating the sets. In order to have more comprehensive logging which can be integrated with other pipelines which may wrap the Venn diagram code, we now use Futile Logger to log the parameters and sets of the Venn diagrams that are generated at runtime [11]. Finally, users can now choose file types of tiff, png or svg, and can alternatively choose to not output a file, but instead output a list of R graphical objects which compose the entirety of the plot. The user can then modify and re-render the plot as desired.

VennDiagramWeb was written using the R statistical programming language and leverages the VennDiagram R package [4]. It uses the web application server Shiny to create a robust graphical user interface which can execute R code on data and parameters dynamically as they change [12]. Using the Shiny web application server enabled us to create a solution that is composed nearly purely of R from end to end. Using a single language allows for very tight integration of security and error handling functionality. We are able to parse any inputs provided by the user and, using functionality built into the language, inspect those inputs to ensure there is no attempt to inject malicious code. The architecture of the web application is based around the code for the user interface and the code for the server. The user interface is defined by a series of widgets which accept the parameters and data files from the user, and display the rendered plot reactively as elements are changed. The server handles all arguments and data, ensures that they are safe and valid, and performs generation of figures.

VennDiagramWeb provides a unique way to integrate and optimize R code for data-visualization with web-based real-time manual optimization with computational pipelines. It will be a key resource for the field. For application support please contact Paul.Boutros@oicr.on.ca.