[Paper Review] BiMAT: a MATLAB(R) package to facilitate the analysis and visualization of bipartite networks

César O. Flores, Timothée Poisot+2 more|arXiv (Cornell University)|Jun 25, 2014

Plant and animal studies34 references3 citations

TL;DR

BiMAT is an open-source MATLAB® package designed to analyze and visualize bipartite ecological networks, implementing algorithms for modularity, nestedness, and multi-scale structure using null models to assess statistical significance. It enables researchers to quantify network structure and visualize results in matrix or graph layouts, supporting ecological and biological network studies with integrated, user-friendly tools.

ABSTRACT

The statistical analysis of the structure of bipartite ecological networks has increased in importance in recent years. Yet, both algorithms and software packages for the analysis of network structure focus on properties of unipartite networks. In response, we describe BiMAT, an object-oriented MATLAB package for the study of the structure of bipartite ecological networks. BiMAT can analyze the structure of networks, including features such as modularity and nestedness, using a selection of widely-adopted algorithms. BiMAT also includes a variety of null models for evaluating the statistical significance of network properties. BiMAT is capable of performing multi-scale analysis of structure - a potential (and under-examined) feature of many biological networks. Finally, BiMAT relies on the graphics capabilities of MATLAB to enable the visualization of the statistical structure of bipartite networks in either matrix or graph layout representations. BiMAT is available as an open-source package at http://ecotheory.biology.gatech.edu/cflores.

Motivation & Objective

• Address the lack of comprehensive, self-contained software tools for analyzing bipartite network structure in ecological and biological systems.
• Provide a unified, extensible MATLAB®-based library that supports both analysis and visualization of bipartite networks.
• Enable statistical significance testing of network features using multiple null models.
• Facilitate multi-scale analysis of network structure, including internal modularity and nestedness at subnetwork levels.
• Support visualization of network structure in both matrix and graph layouts for improved interpretability.

Proposed method

• Implement object-oriented MATLAB® code to represent bipartite networks as adjacency matrices with distinct row and column nodes.
• Integrate widely used algorithms for measuring modularity (e.g., BRIM, adaptive BRIM, leading eigenvector) and nestedness (NTC, NODF).
• Apply null models to evaluate the statistical significance of observed network structure by comparing against randomized network ensembles.
• Automate multi-scale analysis by assessing structural metrics (e.g., Qb, NTC, NODF) at the whole-network and subnetwork levels.
• Utilize MATLAB’s graphics engine to generate interactive and publication-quality visualizations in matrix and graph layouts.
• Support automated sorting of network matrices based on modularity or nestedness to reveal hidden structural patterns.

Experimental results

Research questions

• RQ1 How can bipartite network structure—particularly modularity and nestedness—be quantitatively measured and statistically assessed in ecological systems?
• RQ2 To what extent do network features such as modularity and nestedness co-occur, and how can their interdependence be evaluated?
• RQ3 Can multi-scale analysis reveal structural patterns that are obscured at the whole-network level?
• RQ4 How do geographical or ecological factors influence the formation of modules in host-parasite or virus-host networks?
• RQ5 What role does visualization in matrix versus graph layout play in uncovering structural features like nestedness or modularity?

Key findings

• BiMAT successfully identifies nested and modular structures in empirical plant-pollinator networks, such as Memmot’s network, only when nodes are sorted according to nestedness or modularity.
• The package detects that 12 modules in a virus-host network are significantly less diverse than expected by chance (p = 0.5, one-tailed test), suggesting geographical clustering of phages.
• Multi-scale analysis reveals that internal nestedness (NNTC) values deviate significantly from random expectations, with error bars covering 95% of random replicates.
• Visualization tools in BiMAT clearly distinguish structural patterns: unsorted matrices show no apparent structure, while sorted matrices reveal clear nested and modular patterns.
• The implementation of null models enables robust statistical inference, confirming that observed network features are not due to random chance.
• BiMAT’s extensibility and integration with MATLAB’s visualization tools make it suitable for both exploratory analysis and publication-ready figures in ecological research.