mGEM: multigraph estimation models for pattern analysis

Area of Science:

• Network analysis
• Computational biology
• Systems biology

Background:

• Traditional network analysis often simplifies complex edge data.
• Biological networks, like gene co-expression networks, require advanced analytical tools.
• Existing methods may not fully capture the nuances of multivariate network data.

Purpose of the Study:

• Introduce Multigraph Estimation Models (mGEM) as a flexible alternative to correlation-based methods.
• Develop tools for multivariate analysis of node and edge data in biological networks.
• Improve the interrogation of co-occurrence and co-expression data for enrichment and differential interactions.

Main Methods:

• mGEM models edge weights as multiple edges in a multigraph, using a propensity-based relation.
• A unified parametric framework addresses overdispersion and node-specific attributes.
• Residuals are computed using a background model for ranking node associations and identifying dependent components.

Main Results:

• mGEM successfully analyzes simulated and real-world network data, including neural, co-authorship, and gene co-expression networks.
• Co-authorship network analysis revealed known collaborations and interdisciplinary connections.
• Gene co-expression analysis identified diverse gene modules with unique co-expression propensities.

Conclusions:

• mGEM provides a powerful framework for exploratory data analysis in network science.
• The models generate testable hypotheses from complex network data.
• mGEM offers a more nuanced understanding of biological network structures compared to traditional methods.