A Resampling-Based Framework for Network Structure Learning in High-Dimensional Data.

Area of Science:

• Computational biology
• Network science
• Statistical inference

Background:

• High-dimensional data analysis often faces challenges with limited sample sizes.
• Robust and interpretable network inference is crucial for understanding complex biological systems.
• Existing methods may struggle with mixed data types and higher-order network structures.

Purpose of the Study:

• To introduce RSNet, an open-source R package for resampling-based network inference.
• To provide a framework for robust and interpretable network analysis in high-dimensional settings.
• To address challenges associated with limited sample sizes and mixed data types.

Main Methods:

• Developed a resampling-based framework supporting bootstrap, subsampling, and cluster-based strategies.
• Implemented estimation for partial correlation networks (Gaussian networks) and conditional Gaussian Bayesian networks.
• Integrated graphlet-based topology analysis for higher-order connectivity and edge sign insights.
• Enabled efficient construction of signed graphlet degree vector matrices (GDVMs) for sparse networks.

Main Results:

• RSNet provides statistically reliable network inference for high-dimensional data.
• The package supports mixed data types (continuous and discrete variables).
• Efficient GDVM construction allows scalable analysis of higher-order network structures.
• Graphlet analysis enhances interpretability at single-node and subnetwork levels.

Conclusions:

• RSNet offers a versatile and statistically sound tool for network inference.
• The package enhances interpretability through graphlet-based topology analysis.
• RSNet effectively addresses limited sample size challenges in high-dimensional data analysis.