Nonparametric Bayes Differential Analysis of Multigroup DNA Methylation Data.

Area of Science:

• Genomics
• Biostatistics
• Cancer Research

Background:

• DNA methylation datasets in cancer studies involve numerous cytosine-phosphate-guanine (CpG) sites with complex correlations.
• Identifying disease genomic signatures across different patient groups is a key challenge in cancer research.

Purpose of the Study:

• To propose BayesDiff, a novel nonparametric Bayesian approach for differential DNA methylation analysis.
• To develop a method that can flexibly utilize information from all CpG sites and accommodate serial dependence.

Main Methods:

• Proposed BayesDiff, a nonparametric Bayesian method utilizing a novel class of first-order mixture models (Sticky Pitman-Yor process or 2R2CF).
• Employed simulation studies to compare BayesDiff's effectiveness against existing statistical techniques.
• Applied the methodology to a gastrointestinal (GI) cancer dataset with serial correlation.

Main Results:

• BayesDiff demonstrated effectiveness in identifying differential genomic signatures across patient groups.
• The method adaptively accommodates serial dependence common in DNA methylation data.
• Analysis of a GI cancer dataset supported known aspects of DNA methylation and gene association.

Conclusions:

• BayesDiff offers a robust approach for differential DNA methylation analysis in cancer studies.
• The methodology enhances the identification of disease genomic signatures, particularly in datasets with complex correlation structures.
• Findings complement existing knowledge on DNA methylation and gene association in upper GI cancers.