HiCImpute: A Bayesian hierarchical model for identifying structural zeros and enhancing single cell Hi-C data.

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Single-cell Hi-C (scHi-C) data are crucial for studying cell-to-cell variability in chromatin interactions.
• scHi-C data are severely affected by sparsity (excess zeros) due to low sequencing depth.
• Distinguishing true non-interactions (structural zeros) from sequencing dropouts is vital for accurate downstream analysis.

Purpose of the Study:

• To introduce HiCImpute, a Bayesian hierarchical model designed to identify structural zeros and impute dropout values in scHi-C data.
• To improve the accuracy of downstream analyses, including cell clustering and subtype discovery.

Main Methods:

• Developed HiCImpute, a Bayesian hierarchical model accounting for spatial dependencies in scHi-C data.
• Incorporated information from similar single cells and bulk Hi-C data when available.
• Evaluated HiCImpute using synthetic and real scHi-C datasets.

Main Results:

• HiCImpute accurately identifies structural zeros with high sensitivity.
• The model effectively imputes dropout values in sparse scHi-C data.
• Data processed with HiCImpute led to more accurate cell type clustering compared to existing methods.
• HiCImpute facilitated the identification of novel subtypes within excitatory neuronal cells.

Conclusions:

• HiCImpute provides a robust solution for addressing sparsity in scHi-C data.
• The method enhances the reliability of chromatin interaction analyses and cell-type characterization.
• HiCImpute enables more precise biological discoveries, such as identifying neuronal subtypes.