stabJGL

stabJGL implements a stability-based penalty parameter selection to jointly reconstruct multiple Gaussian graphical models from high-dimensional omics data by selecting network sparsity and between-network similarity.

Key Features:

• Stability-Based Penalty Parameter Selection: Integrates model stability with likelihood-based similarity selection for tuning penalty parameters and addresses limitations of AIC-based criteria in high-dimensional settings.
• Joint Graphical Lasso Enhancement: Builds upon the joint graphical lasso framework to jointly estimate multiple inverse covariance matrices with tuned sparsity and similarity.
• Improved Performance: Demonstrates superior performance relative to the standard joint graphical lasso and other state-of-the-art methods across various performance metrics in high-dimensional data.

Scientific Applications:

• Network Modeling in Omics: Infers gene and protein association networks using Gaussian graphical models from the non-zero entries of inverse covariance matrices.
• Integrative Analysis of Multiple Conditions: Leverages similarities across multiple graphical structures for integrative analyses of complex high-dimensional biological datasets.
• Proteomic Pan-Cancer Analysis: Has been applied to a pan-cancer proteomic study to facilitate discovery of proteomic insights across cancer types.

Methodology:

Extends the joint graphical lasso by incorporating a stability approach to penalty parameter selection that integrates model stability with likelihood-based similarity selection and infers networks via Gaussian graphical models using non-zero entries of inverse covariance matrices.