CpGFilter: model-based CpG probe filtering with replicates for epigenome-wide association studies.

Area of Science:

• Genomics
• Epigenetics
• Statistical Genetics

Background:

• The Infinium HumanMethylation450 BeadChip enables cost-effective epigenome-wide association studies (EWAS).
• A significant portion of CpG sites on the 450K array exhibit substantial measurement errors.
• Noisy CpG sites reduce statistical power in EWAS due to multiple testing corrections.

Purpose of the Study:

• To propose a novel method for filtering unreliable CpG sites in EWAS data.
• To enhance the statistical power and accuracy of EWAS by removing noisy data.
• To develop an efficient computational approach for CpG site filtering.

Main Methods:

• Utilized the intra-class correlation coefficient (ICC) to quantify the ratio of biological variability to total variability.
• Estimated ICC using a linear mixed-effects model, incorporating all samples for robust estimation.
• Developed an ultra-fast algorithm for computationally intensive CpG filtering, enabling rapid analysis of large datasets.

Main Results:

• The proposed whole-sample ICC method demonstrated superior performance compared to replicate-sample ICC and variance-based methods.
• CpG filtering using the ICC method can be completed rapidly on standard hardware for large-scale EWAS datasets.
• The method effectively filters noisy CpG sites, thereby increasing the statistical power of EWAS.

Conclusions:

• The developed ICC-based method provides an effective and efficient approach for filtering noisy CpG sites in EWAS.
• This method improves the reliability and statistical power of EWAS, leading to more accurate identification of relevant genetic associations.
• The CpGFilter R package offers a flexible and accessible tool for researchers conducting EWAS.