Recall DNA methylation levels at low coverage sites using a CNN model in WGBS.

Area of Science:

• Epigenetics
• Genomics
• Computational Biology

Background:

• DNA methylation regulates gene transcription and is quantitatively measured by whole-genome bisulfite sequencing (WGBS).
• WGBS requires high sequencing depth, leading to insufficient coverage at many CpG sites and inaccurate methylation level estimations.
• Existing computational methods often require additional omics data or cross-sample information and may only predict methylation states.

Purpose of the Study:

• To develop a novel computational method, RcWGBS, for imputing missing or low-coverage DNA methylation values in WGBS data.
• To leverage DNA methylation levels from adjacent sites for accurate prediction using deep learning.

Main Methods:

• Proposed RcWGBS, a deep learning-based approach to impute DNA methylation levels.
• Utilized down-sampled WGBS datasets from H1-hESC and GM12878 cell lines.
• Compared RcWGBS performance against METHimpute at low sequencing depths (e.g., 12×).

Main Results:

• RcWGBS achieved high accuracy, with average differences less than 0.03 and 0.01 compared to high-depth data in H1-hESC and GM12878 cells, respectively.
• RcWGBS outperformed METHimpute even at a low sequencing depth of 12×.
• The method effectively imputes DNA methylation levels using information from neighboring sites.

Conclusions:

• RcWGBS provides an accurate and efficient solution for processing low-depth WGBS data.
• This approach can significantly reduce sequencing costs and improve the utility of methylation data.
• Facilitates broader research applications by enabling reliable analysis of epigenomic data with limited sequencing depth.