Bayesian inference of sample-specific coexpression networks.

Area of Science:

• Genomics
• Systems Biology
• Bioinformatics

Background:

• Gene regulatory networks (GRNs) model molecular interactions crucial for biological processes.
• Current coexpression network inference methods often overlook individual variations, providing only population-average insights.
• Understanding heterogeneity in gene regulation is key to deciphering biological system drivers.

Purpose of the Study:

• To introduce Bayesian optimized networks obtained by assimilating omic data (BONOBO), a novel method for inferring sample-specific coexpression matrices.
• To address the limitation of existing methods in capturing population heterogeneity in gene regulatory networks.
• To demonstrate BONOBO's effectiveness in analyzing individual differences in molecular interactions.

Main Methods:

• BONOBO employs a scalable Bayesian model assuming Gaussian distribution on log-transformed gene expression.
• It utilizes a conjugate prior distribution for sample-specific coexpression matrices.
• A closed-form solution for the posterior distribution of coexpression matrices enables efficient analysis of large datasets.

Main Results:

• BONOBO successfully infers sample-specific coexpression matrices, capturing individual variations in molecular interactions.
• The method was validated across diverse datasets, including yeast, human breast cancer, and human thyroid tissue.
• BONOBO demonstrated superior performance compared to existing methods for sample-specific coexpression network inference.

Conclusions:

• BONOBO provides a powerful and scalable approach for inferring individual-specific gene regulatory networks.
• The method offers valuable insights into the drivers of biological processes by accounting for population heterogeneity.
• BONOBO advances the analysis of omics data by revealing inter-individual differences in gene regulation.