Information enhanced model selection for Gaussian graphical model with application to metabolomic data.

Area of Science:

• Computational Biology
• Bioinformatics
• Network Analysis

Background:

• Large biological datasets often suffer from low signal-to-noise ratios, complicating association network analysis.
• Existing methods may struggle to accurately infer network structures from high-dimensional, noisy biological data.

Purpose of the Study:

• To develop a novel method for learning association network structures using Gaussian graphical models (GGMs) integrated with prior knowledge.
• To introduce a new model selection criterion and a data-driven algorithm for constructing candidate models in GGM analysis.

Main Methods:

• Proposed a Structural Bayesian Information Criterion (SBIC) that incorporates prior biological structure into model selection.
• Developed a two-step algorithm to automatically embed prior structure into candidate models for GGM construction.
• Theoretically proved the consistency of SBIC for high-dimensional GGMs under mild conditions.

Main Results:

• The proposed SBIC is a generalization of the extended Bayesian information criterion.
• Simulation studies demonstrated the superiority and robustness of the proposed algorithm compared to existing methods.
• Application to infant fecal metabolite data confirmed the significance of metabolic pathways in metabolite conditional dependence.

Conclusions:

• The novel method effectively learns association network structures from noisy biological data.
• The approach successfully identified new, biologically relevant relationships among metabolites, some previously unrecognized.
• This work provides a powerful tool for dissecting complex biological systems and discovering novel molecular interactions.