Bayesian genome- and epigenome-wide association studies with gene level dependence

Area of Science:

• Genomics
• Biostatistics
• Computational Biology

Background:

• High-throughput genetic and epigenetic data analysis often involves screening numerous variables for phenotype associations.
• Current standard practice uses independent screening with universal multiplicity correction, which may not be optimal for structured genomic data.
• Genomic variables, such as genetic variants and DNA methylation sites, can be logically grouped by gene.

Purpose of the Study:

• To propose a Bayesian hierarchical model for association screening that incorporates gene-specific prior probabilities.
• To provide a more appropriate adjustment for gene- and genome-wide multiplicity in high-throughput genomic studies.
• To demonstrate the application of this method in analyzing DNA methylation differences in The Cancer Genome Atlas (TCGA) glioma data.

Main Methods:

• Developed a Bayesian hierarchical model where prior association probabilities are gene-dependent.
• Modeled gene-specific probabilities nonparametrically within the hierarchical framework.
• Integrated the proposed method into existing Bayesian association screening frameworks.

Main Results:

• The hierarchical Bayesian model allows for refined multiplicity adjustments at both gene and genome levels.
• The approach was successfully applied to screen for differential DNA methylation between lower grade glioma and glioblastoma multiforme.
• The proposed methodology introduces negligible computational overhead.

Conclusions:

• The proposed Bayesian approach offers an effective alternative to standard multiplicity correction methods for high-throughput genomic data.
• This method enhances the accuracy of association screening by leveraging biological structure (gene grouping).
• Available R package `BayesianScreening` facilitates the implementation of this advanced statistical technique.