MambaCpG: an accurate model for single-cell DNA methylation status imputation using mamba.

Area of Science:

• Epigenetics
• Genomics
• Computational Biology

Background:

• DNA methylation is a crucial epigenetic modification influencing biological processes and diseases.
• Accurate DNA methylation analysis is vital for understanding cellular function and pathology.
• Single-cell sequencing advances are hampered by data sparsity, particularly low cytosine-phosphate-guanine (CpG) coverage.

Purpose of the Study:

• To introduce MambaCpG, a novel model for imputing single-cell DNA methylation states.
• To address the challenge of data sparsity in single-cell DNA methylation analysis.
• To leverage the Mamba block architecture for improved methylation pattern recognition.

Main Methods:

• Developed MambaCpG, a single-cell DNA methylation imputation model utilizing the Mamba block.
• Integrated methylation matrix and DNA sequence context for model input.
• Employed bidirectional Mamba blocks to capture long-range dependencies between CpG sites.

Main Results:

• MambaCpG demonstrated superior performance over existing models on large, sparse single-cell DNA methylation datasets.
• Achieved competitive results on smaller datasets.
• Exhibited lower parameter and memory requirements, enhancing practical applicability.
• Identified ultra-long-range dependencies in DNA methylation patterns.

Conclusions:

• MambaCpG effectively imputes sparse single-cell DNA methylation data, enhancing analytical capabilities.
• The model provides novel insights into DNA methylation patterning by capturing long-range dependencies.
• MambaCpG offers a computationally efficient and accurate solution for single-cell epigenomic studies.