Pseudo bulk Single Cell Data

Pseudo-Bulk Single Cell Data Analysis with dcHiC

Single-cell data continues to rise in popularity. With continually improving assays yield datasets with ever greater read depth and tissue-specific heterogeneity, single-cell 3D genome studies have revealed new, more granular insights from developmental dynamics to oncogenesis.

With the proliferation of these datasets, we extended and tested dcHiC for pseudo-bulk single-cell Hi-C data. To use it for analysis,

1. Cluster single cell Hi-C data by your method of choice: usual candidates include by day, by tissue, or by UMAP/t-SNE cluster
2. Combine the individual scHi-C maps for each cluster to get a cluster-wise pseudo-bulk Hi-C map
3. Create an input.txt file, similar to normal analysis but where each row represents a cluster
4. Run dcHiC as usual

Example

We ran dcHiC on postnatal neuronal mice data from Tan et. al's 2021 study In this paper, the author introduced a new scHi-C protocol (Dip-C) and created scHi-C maps for the cortex and hippocampus at days 1, 7, 28, 56, 309, and 347. We separated these into early (1/7), middle (28/56), and late (309/347) stage maps for each cell type -- six different clusters in total. Although scHi-C genomic compartment identification is still difficult, especially given the sparsity of scHi-C data, we were able to analyze and perform differential analysis on profiles with just 100-250 cells per condition.

To perform differential analysis, we first categorized the 6 time points in the study from 2 brain regions into three groups, namely – Early, Mid, and Late. The two time points in each group were treated as replicates; we then applied dcHiC on the aggregated single cell Hi-C maps of each day at 250Kb resolution to call genomic compartments followed by differential analysis among three groups from 2 brain regions separately to identify the compartmental changes below an FDR threshold of 10%.

To see the results, please see the bottom of our visualization page: https://ay-lab.github.io/dcHiC/.