[Paper Review] Graph Models for Biological Pathway Visualization: State of the Art and Future Challenges

Hsiang‐Yun Wu, Martin Nöllenburg+1 more|arXiv (Cornell University)|Oct 10, 2021

Bioinformatics and Genomic Networks38 references4 citations

TL;DR

This paper proposes multilayer networks as a unified data structure for biological pathway visualization, integrating heterogeneous graph models like bipartite graphs, hypergraphs, and reaction graphs into a cohesive framework. By analyzing existing pathway modeling and visualization techniques, it identifies key challenges in scalability, interactivity, time-dependency, uncertainty, and cross-species comparison, offering guidelines for future visualization systems that support dynamic, domain-specific, and cognitively efficient pathway exploration.

ABSTRACT

The concept of multilayer networks has become recently integrated into complex systems modeling since it encapsulates a very general concept of complex relationships. Biological pathways are an example of complex real-world networks, where vertices represent biological entities, and edges indicate the underlying connectivity. For this reason, using multilayer networks to model biological knowledge allows us to formally cover essential properties and theories in the field, which also raises challenges in visualization. This is because, in the early days of pathway visualization research, only restricted types of graphs, such as simple graphs, clustered graphs, and others were adopted. In this paper, we revisit a heterogeneous definition of biological networks and aim to provide an overview to see the gaps between data modeling and visual representation. The contribution will, therefore, lie in providing guidelines and challenges of using multilayer networks as a unified data structure for the biological pathway visualization.

Motivation & Objective

• To address the growing complexity of biological pathways by unifying diverse graph models under a multilayer network framework.
• To identify the disconnect between current data modeling practices and visualization techniques in pathway representation.
• To highlight key challenges in visualizing large-scale, dynamic, and uncertain biological networks.
• To support domain-specific visualization by integrating semantic layers such as function, species, and subcellular compartment.
• To enable comparative analysis across organisms and cellular contexts through a standardized multilayer representation.

Proposed method

• Surveying and classifying underlying graph models used in major pathway databases and visualization tools.
• Mapping conventional biological pathway representations (e.g., substrate graphs, hypergraphs, reaction graphs) to multilayer network abstractions.
• Analyzing hand-crafted and algorithmically generated pathway layouts to assess visual clarity and structural fidelity.
• Evaluating the role of visual hierarchy and layout balance in reducing cognitive load during pathway interpretation.
• Proposing a framework where multilayer networks serve as a central data structure for translating between different visualization types.
• Integrating time, uncertainty, and cross-organism comparisons into a single multilayer modeling and visualization pipeline.

Experimental results

Research questions

• RQ1 How can multilayer networks unify diverse graph models used in biological pathway modeling?
• RQ2 What are the key visualization challenges in representing large-scale, dynamic, and heterogeneous biological pathways?
• RQ3 How can multilayer networks support scalable and interactive visualization of complex metabolic and signaling networks?
• RQ4 In what ways can multilayer networks facilitate comparative pathway analysis across species and cellular compartments?
• RQ5 How can uncertainty and time-dependency be formally modeled and visually represented within a multilayer network framework?

Key findings

• Multilayer networks provide a unifying abstraction that can encapsulate diverse biological pathway models, including bipartite graphs, hypergraphs, and reaction graphs.
• Current visualization techniques often fail to scale with the complexity and density of real biological networks, especially when multiple layers are involved.
• The integration of semantic layers—such as functional modules, species, and subcellular compartments—improves interpretability and supports domain-specific analysis.
• Dynamic and time-dependent pathway changes, such as those in glucose metabolism, require adaptive layout algorithms that are currently underdeveloped.
• Uncertainty in biological interactions is rarely visualized, despite its importance for drug development and systems biology interpretation.
• Comparative pathway visualization across species is facilitated by multilayer networks, which standardize heterogeneous data into a common structural framework.