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

What is it about?

Single cell Hi-C techniques enable one to study cell to cell variability in chromatin interactions, which has significant implications in gene regulations. However, due to insufficient sequencing depth, chromatin interactions low in frequencies may not be observed, leading to many zeros, known as dropouts. There are also zeros due to biological mechanisms rather than insufficient coverage, referred to as structural zeros. As such, dropouts and structural zeros are confounded; that is, observed zeros are a mixture of both types. Differentiating between structural zeros and dropouts is important for improved downstream analyses, including cells-subtype discovery. However, there is a paucity of available methods. In this paper, we develop a powerful method, HiCImpute, for identifying structural zeros and imputing dropouts. Through an extensive simulation study, we demonstrate the ability of HiCImpute for identifying structural zeros with high sensitivity and accurate imputation of dropout values, under a variety of settings. Applications of HiCImpute to three datasets yield improved data that lead to more accurate clustering of cell types, and further, discovery of subtypes in two of the cell types in the prefrontal cortex data.

Featured Image

Photo by Sangharsh Lohakare on Unsplash

Photo by Sangharsh Lohakare on Unsplash