Skip to main content
Fig. 7 | Genome Biology

Fig. 7

From: Normalization and de-noising of single-cell Hi-C data with BandNorm and scVI-3D

Fig. 7

Evaluation of scHi-C normalization and de-noising methods for their impact on downstream analysis. a Percentages of TAD boundaries from Kim2020 data set that are within 1Mb distance of the TAD boundaries of the same cell type bulk Hi-C data. The numbers above each pairwise comparison with BandNorm are the p values based on the two-sided t-test. b HiCRep [13] similarity scores between aggregated scHi-C matrices from the Kim2020 data set and the corresponding bulk Hi-C matrices. The numbers above each pairwise comparison with BandNorm are the p values based on the two-sided t-test. c HiCRep [13] similarity scores between aggregated IMR90 matrices and aggregated GM12878, H1Esc, HAP1, and HFF from Kim2020 data set. The numbers above each pairwise comparison with bulk are the p values based on the two-sided t-test. Sample sizes for a–c are n = 23 for each violin plot corresponding to chromosome 1-22 and chromosome X. d Hierarchical clustering of the aggregated scHi-C matrices of different cell types from BandNorm, scVI-3D, Higashi, and scHiCluster based on their pairwise HiCRep similarity scores, depicted in the heatmap matrices for Lee2019 data set. e Percentage of top N (N = 5,000, 10,000, ...) significant interacting locus pairs that are in the gold standard set for each method on Lee2019 data set. The gold standard set is defined as the top 50,000 significant locus pairs detected by Fit-Hi-C [37] from the cell-type-specific bulk Hi-C data

Back to article page