Fast matrix completion in epigenetic methylation studies with informative covariates

Area of Science:

• Epigenetics and Genomics
• Bioinformatics and Computational Biology

Background:

• DNA methylation is a key epigenetic regulator of gene expression.
• Missing values in omics data, like DNA methylation, pose challenges for analysis and can lead to sample size reduction.
• Integrating high-density (whole genome bisulfite sequencing - WGBS) and low-density (array-based) methylation data is cost-effective but requires robust imputation.

Purpose of the Study:

• To develop and validate an efficient imputation method for DNA methylation data.
• To assess the impact of informative covariates on imputation accuracy.
• To address the common scenario in methylation analysis where the number of features (sites) greatly exceeds the number of samples.

Main Methods:

• Proposed an efficient Linear Model of Coregionalisation with informative Covariates (LMCC).
• The LMCC model incorporates fixed factors (covariates) and latent factors, utilizing Gaussian processes to model spatial correlation.
• Simulated data and applied the method to two real methylation datasets, comparing it with alternative approaches.

Main Results:

• The LMCC model significantly improves imputation accuracy, especially when missing data correlates with covariates.
• The model is efficient for high-dimensional data (more columns than rows), typical in methylation studies.
• Covariates such as cell type, tissue type, and age demonstrably enhance imputation accuracy in real datasets.

Conclusions:

• The proposed LMCC method offers an effective solution for imputing missing DNA methylation values.
• Incorporating relevant covariates is crucial for maximizing imputation accuracy and data utility.
• This approach facilitates better utilization of combined high- and low-density methylation data, enhancing downstream analyses.