gbdmr

gbdmr applies generalized beta regression to detect differentially methylated regions (DMRs) from DNA methylation data by modeling CpG methylation levels with a generalized beta distribution to capture variability and correlation among neighboring CpG sites for phenotype association studies.

Key Features:

• Generalized beta regression: Models DNA methylation levels at CpG sites using a generalized beta distribution rather than linear regression.
• Correlation handling: Explicitly accounts for moderate-to-strong correlation between neighboring CpG sites when detecting DMRs.
• DMR detection: Identifies differentially methylated regions associated with phenotypes rather than relying solely on single-CpG signals.
• Comparative performance: Demonstrated superior DMR discovery compared with dmrff and single-CpG epigenome-wide association studies (EWAS) in scenarios with correlated CpGs.
• Validation on empirical data: Evaluated using simulations and real DNA methylation datasets including the Isle of Wight birth cohort and datasets from the Gene Expression Omnibus.

Scientific Applications:

• Differential methylation analysis: Detects DMRs associated with phenotypes in epigenetic studies.
• Epigenetic epidemiology: Applied to cohort analyses such as the Isle of Wight birth cohort and public GEO datasets for population-level methylation studies.
• Biomarker discovery for precision medicine: Supports identification of methylation regions potentially relevant for disease prediction and phenotype association.

Methodology:

Uses generalized beta regression assuming methylation levels follow a generalized beta distribution, with performance assessed via simulations and analyses of real DNA methylation datasets and comparisons to dmrff and EWAS.