A statistical framework for Illumina DNA methylation arrays.

Area of Science:

• Epigenetics
• Genomics
• Bioinformatics

Background:

• Illumina BeadArray is widely used for DNA methylation profiling.
• Current quality control methods use arbitrary P-value cutoffs, leading to data loss.
• A need exists for methods that utilize all data while accounting for varying observation quality.

Purpose of the Study:

• To develop a statistical framework for bias correction in Illumina Methylation BeadArray data.
• To introduce LumiWCluster, a weighted model-based clustering approach for improved quality control and analysis.
• To enable discovery of distinct methylation patterns and automatic selection of informative CpG loci.

Main Methods:

• Developed a statistical framework using positive control samples for bias removal.
• Introduced LumiWCluster, a weighted model-based clustering algorithm.
• Weighted observations based on detection P-values to account for data quality.

Main Results:

• LumiWCluster systematically weights observations, avoiding data exclusion.
• The method facilitates the discovery of distinct DNA methylation patterns.
• Demonstrated advantages on ovarian cancer and hepatocellular carcinoma datasets.

Conclusions:

• LumiWCluster offers an improved approach to DNA methylation data analysis.
• The method enhances quality control by incorporating all data with differential weighting.
• It aids in identifying biologically relevant methylation patterns and CpG loci.