CompositeView: A Network-Based Visualization Tool

Abstract

Large networks are quintessential to bioinformatics, knowledge graphs, social network analysis, and graph-based learning. CompositeView is a Python-based open-source application that improves interactive complex network visualization and extraction of actionable insight. CompositeView utilizes specifically formatted input data to calculate composite scores and display them using the Cytoscape component of Dash. Composite scores are defined representations of smaller sets of conceptually similar data that, when combined, generate a single score to reduce information overload. Visualized interactive results are user-refined via filtering elements such as node value and edge weight sliders and graph manipulation options (e.g., node color and layout spread). The primary difference between CompositeView and other network visualization tools is its ability to auto-calculate and auto-update composite scores as the user interactively filters or aggregates data. CompositeView was developed to visualize network relevance rankings, but it performs well with non-network data. Three disparate CompositeView use cases are shown: relevance rankings from SemNet 2.0, an open-source knowledge graph relationship ranking software for biomedical literature-based discovery; Human Development Index (HDI) data; and the Framingham cardiovascular study. CompositeView was stress tested to construct reference benchmarks that define breadth and size of data effectively visualized. Finally, CompositeView is compared to Excel, Tableau, Cytoscape, neo4j, NodeXL, and Gephi.

Keywords: CompositeView; HeteSim; SemNet; biomedical knowledge graph; concept relatedness; link prediction; network analysis.

Figure 1.

A high-level overview of CompositeView’s working cycle. CompositeView has placeholder data, which initialize the graph, and user data, which initialize the user interaction and application of CompositeView. The cycle begins with a user uploading data and interacting with the application to update the graph attributes and layout. Next, the Cytoscape elements are updated to run the graph. Finally, the graph rendering and display are visually updated to the user in the CompositeView application. The working cycle continues as the user makes updates to the data or changes or applies CompositeView application features such as graph layout selection or filtering modes.

Figure 2.

The three input data examples explained in Section 2.1.3, evolving from least to most complex.

Figure 3.

A sample of tested graph layouts along with their CompositeView runtimes, all based on the same SemNet results data set (approximately 2472 source nodes).

Figure 4.

The adjusted spring graph layout using the same SemNet test data from Figure 3 (runtime: 5.78 s).

Figure 5.

The adjusted spring layout method, broken down into three logical steps. The data shown are placeholder data used in CompositeView. (a) Initial target nodes are simulated and positions are fixed. (b) Artificial edges are removed and source nodes are filled in around the shared target node centroids. (c) Source nodes are simulated with edge weights.

Figure 6.

Value filtering as well as node and type filtering settings as displayed by CompositeView. (a) Value filtering sliders. (b) Node and edge filtering dropdowns.

Figure 7.

The complete CompositeView application layout (with Graph Manipulation settings open).

Figure 8.

The impact of source spread (k in the NetworkX spring layout). The data shown is placeholder data used in CompositeView. (a) Source spread value half of base. (b) Base source spread value, the same as Figure 5c. (c) Source spread value double of base.

Figure 8.

The impact of source spread (k in the NetworkX spring layout). The data shown is placeholder data used in CompositeView. (a) Source spread value half of base. (b) Base source spread value, the same as Figure 5c. (c) Source spread value double of base.

Figure 9.

Application startup, graph initialization, attribute loading, and graph update.

Figure 10.

Runtime analysis of graph startup and update, broken down by most important methods.

Figure 11.

The SemNet sample data, both unfiltered (a) and filtered (b), based on criteria described in Section 2.3.1.

Figure 12.

The HDI sample data, both unfiltered (a) and filtered (b), based on criteria described in Section 2.3.2.

Figure 13.

The CVD sample data, both unfiltered (a) and filtered (b), based on criteria described in Section 2.3.3.

Figure 14.

Visual comparison between Gephi and CompositeView using the same SemNet data set seen in Figures 3 and 4 (approximately 2472 source nodes). The red circles represent the main input features or nodes for which relationships are being visualized. In this example, visualized relationships for composite data are much easier to deduce with CompositeView compared to Gephi.