GraphCpG: imputation of single-cell methylomes based on locus-aware neighboring subgraphs

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Single-cell DNA methylation sequencing offers high resolution but suffers from incomplete data coverage.
• This sparsity hinders accurate downstream genome-wide analyses.
• Scalable imputation methods are crucial for large-scale single-cell epigenomic studies.

Purpose of the Study:

• To develop a scalable and efficient deep learning approach for imputing sparse single-cell DNA methylation matrices.
• To enhance the accuracy and utility of genome-wide analyses from single-cell methylation data.

Main Methods:

• Proposed a novel graph-based deep learning framework, GraphCpG.
• Utilized locus-aware neighboring subgraphs and locus-aware encoding for imputation.
• Applied the method to CpG methylation matrices.

Main Results:

• GraphCpG outperforms existing methods on datasets with hundreds of cells and shows competitive results on smaller datasets.
• Demonstrated improved imputation performance with increasing cell numbers.
• Significantly reduced computation time and enhanced downstream analysis outcomes.

Conclusions:

• GraphCpG offers a powerful and efficient solution for imputing sparse single-cell DNA methylation data.
• The method's scalability and performance make it suitable for large-scale epigenomic research.
• Improved imputation facilitates more robust downstream analyses in single-cell epigenetics.