Filtering High-Dimensional Methylation Marks With Extremely Small Sample Size: An Application to Gastric Cancer Data

Area of Science:

• Epigenetics
• Genomics
• Bioinformatics

Background:

• DNA methylation is crucial in cancer development and treatment response.
• Identifying significant methylation sites from high-dimensional data is challenging due to scalability and statistical issues.
• A rapid pre-screening method is needed to filter candidate methylation sites for further analysis.

Purpose of the Study:

• To introduce and evaluate BACkPAy, a Bayesian pre-screening approach for differential DNA methylation analysis.
• To identify biologically meaningful patterns of differential methylation in gastric cancer.
• To validate potential prognostic biomarkers for gastric cancer using The Cancer Genome Atlas (TCGA) data.

Main Methods:

• Applied BACkPAy to a genome-wide DNA methylation dataset of gastric cancer samples.
• Compared BACkPAy results with LIMMA (Linear Models for Microarray and RNA-Seq Data) using Benjamin-Hochberg FDR.
• Utilized Cox proportional hazards regression models and TCGA data for survival analysis of identified markers.

Main Results:

• BACkPAy identified eight groups of differential methylation probes from the gastric cancer dataset.
• Five prognostic genes (RDH13, CLDN11, TMTC1, UCHL1, FOXP2) with differential methylation probes were identified using TCGA data.
• LIMMA with Benjamin-Hochberg FDR did not yield significant results, highlighting BACkPAy's efficacy with small sample sizes.

Conclusions:

• BACkPAy is effective for analyzing DNA methylation data, especially with small sample sizes in gastric cancer research.
• The identified genes RDH13, CLDN11, TMTC1, UCHL1, and FOXP2 show potential as predictive biomarkers for gastric cancer treatment.
• Promoter methylation levels of these genes in serum may have prognostic and diagnostic value for gastric cancer patients.