Skip to main content
Fig. 3 | Genome Biology

Fig. 3

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

Fig. 3

Separation of the neuronal sub-cell types in the Lee2019 data set and the impact of the batch effects. a Application of the scHi-C data normalization and de-noising methods on Lee2019 data set with 14 neuronal cell types. The results are displayed using scatter plots of the two UMAP coordinates. The colors of the plotting symbols correspond to the cell types. Excitatory neuronal subtypes (L2/3, L4, L5, L6) and inhibitory cells (Ndnf, Vip, Pvalb, and Sst) are highlighted by the gray dashed squares, which are amplified in b for the four leading methods: BandNorm, scHiCluster, Higashi, and scVI-3D, respectively. c Impact of batch effects on cell type separation using Lee2019 data set with samples from two donors of ages 21 and 29 years old and in a total of 5 batches. The results are displayed using scatter plots of the two UMAP coordinates. The colors of the plotting symbols correspond to the batches. d, e Cell type separation and batch effect removal performances of the methods, evaluated by ARI with K-means clustering (k = number of cell types) in d and Louvain clustering in e. ARI assesses the batch effect with K-means clustering (k = number of batches). f Density of integration local inverse Simpson’s Index (iLISI) scores [25] of cells evaluating the batch mixing performance after normalization by each of the eight methods. A high density around 1 indicates only one batch in a particular cell’s neighborhood, demonstrating the batch effect. Larger values indicate well mixing across the batches

Back to article page