CaMelia

CaMelia imputes missing CpG methylation states in single-cell DNA methylation data to mitigate sparsity and improve downstream epigenetic analyses.

Key Features:

• CatBoost gradient boosting: Uses the CatBoost gradient boosting model to model complex relationships in methylation data.
• CpG-site imputation: Imputes missing methylation states at specific CpG sites within single-cell DNA methylation sequences.
• Intercellular and bulk-data borrowing: Leverages information from similar loci in other cells or from bulk data to inform imputations.
• Locally paired methylation patterns: Exploits locally paired similarities in methylation patterns around target loci to guide predictions.
• Preservation of biological relationships: Preserves cell–cell biological relationships during imputation to support accurate downstream analyses such as clustering.
• Empirical performance: Demonstrated improved performance on real single-cell methylation datasets relative to previous imputation methods.

Scientific Applications:

• Single-cell methylome completion: Enables more complete single-cell DNA methylation datasets for epigenetic analyses.
• Cell-type identification and clustering: Improves cell-type classification and clustering by maintaining cell–cell relationships and revealing differentially methylated loci.
• Detection of differentially methylated loci: Uncovers differentially methylated loci that are obscured by sparse coverage in single-cell data.
• Studies of cellular hierarchies and gene regulation: Supports epigenetic investigations of complex cellular hierarchies and gene regulation mechanisms by increasing methylome resolution.

Methodology:

CaMelia applies CatBoost gradient boosting to predict and impute missing CpG methylation states by analyzing intercellular methylation pattern similarities and using information from similar loci in other cells or bulk data based on locally paired methylation pattern similarities around target loci.