An evaluation of statistical methods for DNA methylation microarray data analysis.

Area of Science:

• Epigenetics
• Genomics
• Bioinformatics

Background:

• DNA methylation is a key epigenetic mechanism regulating gene expression.
• Common quantification methods include β values and M values.
• Various statistical approaches exist for DNA methylation data analysis, but method selection can be challenging.

Purpose of the Study:

• To compare the performance of six statistical methods for DNA methylation microarray analysis.
• To evaluate methods based on false discovery rate control, statistical power, and stability.
• To provide guidance on optimal statistical method selection for different study scenarios.

Main Methods:

• Comparative analysis of six statistical methods (Wilcoxon rank sum test, t-test, Kolmogorov-Smirnov test, permutation test, empirical Bayes, bump hunting).
• Simulation studies and real data examples (ovarian cancer, rheumatoid arthritis) were used.
• Evaluation metrics included false discovery rate (FDR) control, statistical power, and method stability.

Main Results:

• Empirical Bayes and bump hunting methods demonstrated appropriate FDR control and high power for small sample sizes with independent methylation levels.
• Bump hunting method showed appropriate FDR control and high power for small sample sizes with correlated methylation levels.
• For medium to large sample sizes, most methods performed similarly, except for the permutation test with low proportions of differentially methylated loci. Bump hunting showed lower stability with large proportions of differentially methylated loci.

Conclusions:

• For small sample sizes, empirical Bayes and bump hunting are recommended for independent methylation levels; bump hunting is recommended for correlated levels.
• All tested methods are acceptable for medium and large sample sizes.
• The study provides practical guidance for selecting appropriate statistical methods in DNA methylation research.