Mouse obesity network reconstruction with a variational Bayes algorithm to employ aggressive false positive control

Area of Science:

• Genomics
• Systems Biology
• Bioinformatics

Background:

• High-throughput genomic datasets present challenges for identifying disease-related factors.
• Existing network recovery methods often require cross-validation or model selection criteria.
• Understanding genetic and gene expression influences on obesity is crucial.

Purpose of the Study:

• To develop a novel variational Bayes network reconstruction algorithm for extracting relevant disease factors.
• To apply the algorithm to identify genetic markers and liver gene expression traits associated with obesity phenotypes in mice.
• To evaluate the algorithm's performance against existing network recovery methods.

Main Methods:

• A scalable, regularized network recovery algorithm employing Bayesian model averaging.
• Internal estimation of sparsity to minimize false positives without cross-validation.
• Application to an F2 intercross mouse dataset examining obesity-related phenotypes (weight, cholesterol, glucose, free fatty acids).

Main Results:

• Eleven novel genes and one quantitative trait locus directly linked to obesity-related phenotypes were identified.
• These genes, though not found by other analyses, have prior associations with obesity.
• The algorithm demonstrated superior performance in power and type I error control compared to lasso-based methods.

Conclusions:

• The developed network contains 118 associated and novel genes implicated in obesity.
• These genes represent excellent candidates for obesity risk factors.
• The algorithm provides a robust tool for dissecting complex genetic networks related to metabolic diseases.