Project Overview 

This project developed MultiNet – https://multinet.app, a general, robust, and sustainable software tool for reasoning about and analyzing multivariate networks and trees. The completed system allows researchers across multiple disciplines to upload their datasets, curate their data into a connected network structure, and view the resulting network using advanced, interactive visualizations. MultiNet is unique because it visualizes both the structure of the network and the attributes of the nodes and edges, unlike alternative systems that mostly visualize structure and possibly a single attribute. 

MultiNet supports three core network-analysis capabilities: (1) reasoning about relationships using state-of-the-art techniques for interactively visualizing both the attributes and connectivity of large multivariate networks; (2) inferring new relationships using techniques for dynamic, user-driven reshaping of the underlying network structure based on its attributes; (3) reporting findings through provenance mechanisms that capture analysis sessions in support of reproducibility, dissemination, and collaboration. 

The software design of the MultiNet framework uses a plug-in architecture model so the platform can be further extended with other general or domain-specific analysis interfaces in the future. A well-defined programming interface for accessing and updating the stored networks is designed into MultiNet to ease future development.  In addition to the public MultiNet site, we designed the MultiNet system to use the Docker run-time container technology so individual researchers or institutions can instantiate their own version of MultiNet easily to use for private data.   

Outcomes 

The key technical outcomes of MultiNet are: 

1. A platform for securely uploading and sharing network data and to develop network visualization plug-ins against. The platform solves critical infrastructure issues, such as user management, data management, data formatting, data transformation, hosting, etc. 

2. The MultiLink Node-Link visualization; an advanced node-link view that implements sophisticated approaches to integrate node/edge attributes and network structure. Among the key methods to show attributes are node styling, embedding of charts in nodes, and  attribute-driven layouts.

3. The MultiMatrix Adjacency Matrix visualization; an advanced adjacency matrix that supports grouping of nodes by attributes and rich attribute visualizations in a juxtaposed table. 

4. An implementation of UpSet, a  set (hypergraph) visualization tool. Upset visualizes set data as a matrix of set inclusion, instead of with traditional approaches like venn diagrams.  UpSet also visualizes attributes of the intersections.  

We evaluated MultiNet during the course of the project by partnering with experts in the disciplines of phylogenetics, neurology, social networking, and mineralogy. For each use case, our team developed innovative visualizations that streamlined the analysis of the domain-specific networks in each research case.  

Phylogenetic biology focuses on understanding and modeling the rate of evolutionary change and species formation in plant, animal, and microbial communities.  Analytical modeling in phylogenetics often centers around binary or multi-way trees that represent patterns of common ancestry of species in communities, and can be used to understand the dynamics of change over time.  The MultiNet system provides a novel way to view attributed phylogenetic trees. 

With our retinal connectomics collaborator at Utah, we created a robust visual interface for multi-hop querying and exploring the connectivity of large, interconnected neural fabrics. We developed a domain-specific network querying algorithm. Additionally, we implemented an automated workflow that updates their data stored on the MultiNet application. 

In collaboration with computational social scientists at Duke, we developed visualizations to address temporal/dynamic network dataset that reflect the organization of social groups changing over time. The tasks included time varying visualizations and also visual representations of edge distributions of categorical and numerical attributes, thus expanding the visualization contribution beyond dynamic networks. We focused on matrix representations of edge distributions by attributes and representations of sociological constructs (e.g. triadic distributions). MultiNet’s provenance tracking was used to explore dynamic network changes. 

For mineralogy researchers from Carnegie Science, we focused on unipartite and bipartite networks of minerals in selected groups on Earth and in martian meteorites. We customized the node appearance (size, color, shape) to represent various attributes of the mineralogical systems, including chemical composition, frequency of occurrence, and crystal structure. This work provided a holistic view of complex mineral systems and aided in scientific discovery.

Broader Impacts

Our sustainable, modular software design enables others working on multivariate networks to use MultiNet as a platform upon which to build and add capabilities. We have already established various examples of how the MultiNet infrastructure has furthered other research efforts.  The novel capabilities of MultiNet have led to new insights in collaboration application areas, with publications in retinal connectomics and evolutionary biology specifically crediting MultiNet. 

Through their involvement in the MultiNet project, a diverse group of students has gained experience doing collaborative work. The project has supported several PhD students; one of which has already graduated. The project also supported several undergraduate researchers through an REU. Several students have since entered the STEM workforce. 

Last Modified: 05/03/2024
Modified by: Alexander Lex