Single-cell specific and interpretable machine learning models for sparse scChIP-seq data imputation.

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Single-cell Chromatin ImmunoPrecipitation DNA-Sequencing (scChIP-seq) is crucial for understanding cellular heterogeneity.
• Data sparsity is a major challenge in scChIP-seq analysis, limiting insights into protein-DNA interactions.
• Existing imputation methods are not specifically designed for the unique challenges of scChIP-seq data.

Purpose of the Study:

• To introduce SIMPA, a novel imputation method tailored for scChIP-seq data.
• To leverage bulk ENCODE data for imputing missing protein-DNA interactions in single cells.
• To enhance the interpretability of scChIP-seq data analysis.

Main Methods:

• Developed SIMPA, a machine learning-based imputation algorithm for scChIP-seq data.
• Trained imputation models on a per-cell, per-ChIP target, and per-genomic region basis.
• Utilized predictive information from ENCODE bulk ChIP-seq data.
• Incorporated feature importance analysis for model interpretability.

Main Results:

• SIMPA accurately preserves cell type clustering and improves pathway identification in human data.
• Imputed profiles demonstrate single-cell specificity, outperforming generic imputation.
• As few as 100 input regions enable accurate imputation of thousands of undetected regions.
• Feature importance analysis of H3K4me3 profiles correlates with gene co-expression in cell-type specific pathways.

Conclusions:

• SIMPA effectively addresses data sparsity in scChIP-seq, enabling deeper biological insights.
• The method's interpretability provides a mechanistic understanding of imputed interactions.
• SIMPA enhances the utility of scChIP-seq for studying epigenomic regulation at single-cell resolution.