Model for heterogeneous random networks using continuous latent variables and an application to a tree-fungus network

Area of Science:

• Network Science
• Graph Theory
• Statistical Modeling

Background:

• Mixture models are widely used for heterogeneous random graphs, capturing structures like hubs and communities.
• Existing mixture models have limitations in computational feasibility and handling vertices with intermediate positions.
• There is a need for more flexible and theoretically sound models for complex network analysis.

Purpose of the Study:

• To propose a novel grade of membership model for heterogeneous random graphs.
• To address the limitations of existing mixture models in terms of computational feasibility and vertex representation.
• To provide a more nuanced framework for understanding network heterogeneity.

Main Methods:

• Developed a grade of membership model where each vertex is a mixture of extremal hypothetical vertices.
• Connectivity properties are derived from the extreme vertices' properties.
• Employed a maximum likelihood procedure for parameter estimation.
• The model's tractability is ensured by the number of observations being proportional to the square of the network's vertices.

Main Results:

• The proposed model allows vertices to occupy intermediate positions, unlike traditional mixture models.
• Parameter estimation via maximum likelihood is computationally feasible.
• The model successfully elucidates processes shaping heterogeneous host/parasite interaction networks.

Conclusions:

• The grade of membership model offers a significant advancement for analyzing heterogeneous random graphs.
• This new model provides a more flexible and theoretically robust alternative to existing methods.
• The model has practical applications in understanding complex biological interaction networks.