Reverse engineering gene networks using global-local shrinkage rules.

Area of Science:

• Computational Biology
• Systems Biology
• Bioinformatics

Background:

• Inferring gene regulatory networks (GRNs) from high-throughput 'omics' data is crucial but computationally challenging.
• Classical methods struggle with the curse of dimensionality and lack robustness.
• Existing regularized methods often use non-robust loss functions and hinder prior information integration.

Purpose of the Study:

• To develop a robust Bayesian hierarchical model for reconstructing GRNs from time-series gene expression data.
• To address limitations of classical and regularized methods in GRN inference.
• To enable straightforward incorporation of prior biological knowledge.

Main Methods:

• A Bayesian hierarchical model utilizing global-local shrinkage priors for edge selection.
• Relaxation of the normal likelihood assumption to accommodate heavy-tailed gene expression data.
• Development of an efficient Markov chain Monte Carlo (MCMC) algorithm via Gibbs sampling.

Main Results:

• The proposed model demonstrates improved performance in simulation studies for GRN reconstruction.
• The methodology effectively handles heavy-tailed data, a common challenge in gene expression analysis.
• Comparison with existing methods on real T-cell activation data shows competitive or superior performance.

Conclusions:

• The developed Bayesian approach offers a robust and flexible framework for gene regulatory network inference.
• The model's ability to handle heavy-tailed distributions and incorporate prior information enhances its applicability.
• This method provides a significant advancement in reconstructing complex biological networks from high-throughput data.