Extended Data Fig. 2: Genetic network analysis of the HCL.
From: Construction of a human cell landscape at single-cell level
a, Verification of pseudo-cell analysis for network interpretation from HCL data. Left, numbers of genes in single cell, pseudo-cell 5, pseudo-cell 10, pseudo-cell 20, pseudo-cell 50 and pseudo-cell 100 from mouse lung single-cell data14,15 generated by 10X Genomics, Smart-seq2 and microwell-seq. Genes were calculated in each cell or pseudo-cell. Sample sizes for each box from left to right were: 10X Genomics: 5,449, 1,089, 540, 272, 112 and 62; Smart-seq2: 1,620, 324, 158, 83, 33 and 20; microwell-seq: 6,940, 1,390, 686, 349, 142 and 81. Right, silhouette value in single cell, pseudo-cell 5, pseudo-cell 10, pseudo-cell 20, pseudo-cell 50 and pseudo-cell 100 from mouse lung single-cell data. A high silhouette value represents good separation. Sample sizes for each box from left to right were: 6,940, 1,390, 686, 349, 142 and 81. Box plots: centre line, median; boxes, first and third quartiles of the distribution; whiskers, highest and lowest data points within 1.5 × IQR. b, t-SNE map of HCL pseudo-cell data showing improved cell-type clustering (n = 30,053 pseudo-cells). c, TF–TF correlation heat map covering 1,521 human TFs generated using HCL pseudo-cell data. The correlation data are listed in Supplementary Table 2. d, Representative TF network in the HCL (PCC > 0.5). Note that the HCL TF network is highly related within small modules but discrete among different modules.
