Introduction to correlation networks: Interdisciplinary approaches beyond thresholding

Area of Science:

• Interdisciplinary Network Science
• Statistical Modeling
• Data Analysis

Background:

• Empirical networks frequently derive from correlational data across diverse fields like psychology, neuroscience, and finance.
• Specialized network analysis methods exist in various domains, but cross-disciplinary communication is limited.
• Transforming correlation matrices into networks presents challenges, with thresholding being common but problematic.

Purpose of the Study:

• To review and compare various methods for constructing and analyzing correlation networks.
• To highlight limitations of common methods like thresholding.
• To propose best practices and identify open questions in correlation network analysis.

Main Methods:

• Review of existing literature on correlation network construction and analysis.
• Discussion of methods including thresholding, weighted networks, regularization, dynamic networks, and threshold-free approaches.
• Comparison with null models and consideration of unweighted vs. weighted networks.

Main Results:

• Thresholding correlation matrices can lead to suboptimal network representations.
• A variety of advanced techniques exist, offering improvements over basic thresholding.
• No single method is universally superior; the choice depends on the specific application.

Conclusions:

• Cross-disciplinary insights are crucial for advancing correlation network analysis.
• Recommended practices and open research questions are proposed for the field.
• Further research is needed to optimize network construction from correlational data.