setsApp for Cytoscape: Set operations for Cytoscape Nodes and Edges

Introduction

Cytoscape^1,2 provides an environment for the visualization and analysis of networks and associated annotations. The primary audience for Cytoscape is the biological community and Cytoscape supports a number of standard use cases for analyzing and visualizing biological data. Many of these use cases involve the selection of a number of nodes or edges based on some analysis or annotation and either performing an action on that selection or comparing those nodes or edges to a different set of nodes or edges that resulted from alternative analyses or analyses based on alternative annotations. The core capabilities for Cytoscape provide some tools to facilitate these types of comparisons but they can be counterintuitive or complicated to use. setsApp is a Cytoscape 3 application that provides a general set of tools for users and developers to define and maintain sets of nodes or edges and compare those sets using the standard set operations of union, intersection, and difference. Partition and layout features are also provided to assist in generating sets and performing set-aware layouts.

In this paper, we present the implementation of setsApp, in particular, how setsApp integrates with the Cytoscape command system, and then present a sample biological workflow using setsApp.

Implementation

Cytoscape provides two approaches to implementing apps: a simple app and a bundle app. Simple apps are implemented using the same general approach as in Cytoscape 2.8, but at the cost of significant flexibility. Bundle apps utilize Open Service Gateway Initiative (OSGi)³ interfaces through APIs provided by Cytoscape to interact with the Cytoscape core functionality. setsApp is implemented as a bundle app and utilizes the Cytoscape 3.1.0 API. There are three main components to the setsApp implementation: the user interface, the command interface, and the underlying data model for maintaining sets of nodes and edges.

SynchronousTaskManager tm =
  serviceRegistrar. getService
      (SynchronousTaskManager.class);
CommandExecutorTaskFactory cetf =
  serviceRegistrar. getService
      (CommandExecutorTaskFactory.class);
Map<String, Object> argMap = new
    HashMap<String, Object>();
argMap.put("name", "NewSet");
// selected is a special keyword for the
    NodeList tunable
argMap.put("nodeList","selected");
// the current network
argMap.put("network","current");
// Assumes that this implements TaskObserver
tm.execute(cetf.createTaskIterator("setsApp",
    "createSet", argMap, this), this);

Results

A simple example use case might be the exploration of the data set from Ideker et al., 2001⁴, which measured the change in expression for 331 genes after a systematic deletion of genes known to be involved in the Saccharomyces cerevisiae switch to galactose metabolism. This data was combined with known protein-protein and protein-DNA interactions to explore the biological response to deletions in the presence (the G in the column names) or absence of galactose in the medium. This data set is now included as a sample with Cytoscape downloads (galFiltered.cys). In our workflow we use Cytoscape’s Select panel to select all proteins that are underexpressed (gal1RGexp < -0.5 fold change) in the deletion of GAL1 (Figure 1). That selection is saved as a set named GAL1- (Apps→SetsApp→Create node set). We then select all of the proteins that are overexpressed (gal1RGexp > 0.5 fold change) in the deletion of GAL1 and save that selection as a set named GAL1+. Repeating this for GAL4 (column gal4RGexp) and GAL80 (column gal80Rexp) results in 6 sets altogether: GAL1+, GAL1-, GAL4+, GAL4-, GAL80+, GAL80- (Figure 2). Note that the data for GAL1 and GAL4 is in the presence of galactose, but the data for GAL80 is in the absence of galactose since GAL80 is a known repressor of GAL4.

Figure 1. Screenshot of Cytoscape’s Select panel with underexpressed genes in the gal1RGexp condition being selected.

Nodes that match the gal1RGexp condition are highlighted in red.

Figure 2. Screenshot of Cytoscape showing the Sets panel with all condition sets created.

Nodes in the GAL+ set are highlighted in red.

Given those six sets of genes, we can explore the data sets by looking at combinations of the sets. For example, we could look at the intersection of all of the underexpressed proteins by selecting each of GAL1-, GAL4-, and GAL80- in the Sets panel and pressing the Intersection button in the Set Operations box near the bottom of the panel. If we name the resulting set GAL- we see that it contains a single gene: YOL058W (ARG1). In this data set of 331 genes, only this one gene is repressed for all three of the deleted GAL genes. In the absence of galactose when GAL80 is deleted, ARG1 is underexpressed, and in the presence of galactose when either GAL1 or GAL4 are deleted the gene is also underexpressed. Looking at the expression significance values in the Node Table Panel of Cytoscape (gal1RGsig, gal4RGsig, and gal80Rsig) this is a highly significant result, although there is no direct correlation between the galactose switch and arginine biosynthesis regulation that we were able to find in the literature. On the other hand, ARG1 is regulated by the GCN4 activator which is known to repress protein synthesis during periods of stress or starvation⁵, which explains the significant down-regulation of ARG1. We can perform a similar analysis to understand the consistent up-regulation of the five genes in the GAL+ set corresponding to gene symbols: GIP1, NCE103, YIG1, POT1, and ICL1. Figure 2 shows the results of the intersection operations.

We can also explore data sets by using layout algorithms that are informed by the sets we created earlier. By default, Cytoscape provides over a dozen layout algorithms that spatially place nodes in order to help elucidate meaningful patterns of relationships in networks. Most layout algorithms take into account connectivity between nodes. SetsApp augments Cytoscape by supplying two additional layout algorithms that take set membership into account. The setsbased grid layout places nodes in the same set together into independent grids, i.e., a grid of grids. This is ideal for quickly separating nodes in different sets for manual manipulation later on. The setsbased force directed layout employs the Prefuse force directed layout provided by Cytoscape but also tries to put nodes in the same set closer together in the network. Users can adjust the force between nodes in a given set relative to connected nodes and thereby emphasize or diminish the grouping based on set membership by changing values in the layout settings panel (Layout→Settings... ).

Conclusions

There are many Cytoscape workflows that could take advantage of the features of setsApp. Any workflow that might want to look for groups of nodes or edges that share multiple traits, or that explicitly do not share those traits. While it is possible to duplicate many of the final results enabled by setsApp by using Cytoscape 3.1’s new Select panel, a user would need to know in advance exactly the combination of features that were biologically interesting. setsApp provides an alternative that allows users to explore various combinations of nodes and edges and to save such selections for later uses. We also provide layout algorithms that take set membership into consideration.

In the workflow we developed above, we combined the functionality of Cytoscape’s Select panel with setsApp to explore combinations of sets of genes based on shared properties. There are many more sophisticated apps available to Cytoscape users that could be used to do a more thorough analysis of this data set including jActiveModules⁶, clusterMaker⁷ and RINalyzer⁸, however, the workflow above demonstrates the utility of a simple set-oriented approach to exploring networks.

Software availability

Software available from: http://apps.cytoscape.org/apps/setsApp

Latest source code: https://github.com/RBVI/setsApp

Source code as at the time of publication: https://github.com/F1000Research/setsApp

Archived code as at the time of publication: http://www.dx.doi.org/10.5281/zenodo.10424⁹

License: Lesser GNU Public License 3.0: https://www.gnu.org/licenses/lgpl.html