CaMelia: imputation in single-cell methylomes based on local similarities between cells

Area of Science:

• Epigenetics
• Genomics
• Computational Biology

Background:

• Single-cell DNA methylation sequencing offers single-cell resolution for understanding epigenetic regulation of gene expression.
• Current technologies face challenges with low CpG coverage, leading to sparse data.
• Addressing data sparsity is crucial for accurate whole-genome quantitative analysis.

Purpose of the Study:

• To develop a computational method for imputing missing methylation states in sparse single-cell DNA methylation data.
• To enhance downstream analyses such as cell-type identification and subpopulation discovery.

Main Methods:

• Developed CaMelia, a CatBoost gradient boosting algorithm.
• Leveraged locally paired intercellular methylation pattern similarity for imputation.
• Applied the method to real single-cell methylation datasets.

Main Results:

• CaMelia demonstrated significant imputation performance gains compared to existing methods.
• Imputed data facilitated the discovery of more differentially methylated loci masked by sparsity.
• Clustering analysis showed improved cell-type and subpopulation identification, preserving cell-cell relationships.

Conclusions:

• CaMelia effectively addresses data sparsity in single-cell methylation sequencing.
• The method enhances the biological insights obtainable from single-cell epigenomic data.
• CaMelia offers a valuable tool for advancing single-cell epigenetics research.