A varying-coefficient model for the analysis of methylation sequencing data.

Area of Science:

• Epigenetics
• Genomics
• Bioinformatics

Background:

• DNA methylation is a key epigenetic regulator of gene expression.
• Next-generation sequencing provides single-base-resolution DNA methylation data.
• Noisy and sparse data, especially with few replicates, challenge the identification of differentially methylated regions (DMRs).

Purpose of the Study:

• To develop a robust statistical model for detecting DMRs from single-base-resolved DNA methylation data.
• To address challenges posed by noisy, sparse data and limited replicates in DMR identification.
• To incorporate covariate information and model overdispersion in methylation analysis.

Main Methods:

• A varying-coefficient model is proposed for DMR detection.
• The model simultaneously smooths methylation profiles and identifies DMRs.
• A beta-binomial distribution is used to account for overdispersion, which can be modeled based on genomic region and covariates.

Main Results:

• The proposed varying-coefficient model effectively detects DMRs in single-base-resolved methylation data.
• The model demonstrates robustness in handling noisy and sparse datasets, even with few replicates.
• Application to two case studies validates the model's performance and utility.

Conclusions:

• The developed varying-coefficient model offers an effective approach for identifying DMRs.
• This method improves the analysis of DNA methylation data, particularly in scenarios with limited sample sizes and data sparsity.
• The model provides a valuable tool for epigenetic research and gene regulation studies.