Skip to main content
. 2020 Jul 8;6(28):eaba1983. doi: 10.1126/sciadv.aba1983

Fig. 6. GRN analysis of IPF and control lungs.

Fig. 6

(A) Overview of bigSCale method for computing GRN. (i) Cells are recursively clustered down to subclusters. (ii) Z scores are calculated on the basis of differential expression between subclusters (iii) correlations between all genes via Pearson and cosine. (iv) Correlation edges are thresholded using the top 99.3% quantile of correlation coefficients; only edges where at least one node has a Gene Ontology (GO) annotation as a gene regulator. (B) Summary of network structure for control and IPF GRNs. (C) GRNs of control and IPF lung cells. Nodes represent genes, and edges represent correlations of putative regulatory relationships. Nodes sizes correspond to PageRank centralities, and the largest clusters are assigned colors to their nodes, with each color representing a distinct cluster. The top cell types relevant to each highlighted cluster are shown. Behind each highlighted cluster is a polygon shape covering the domain of the cluster colored by the category of cell type that is predominantly relevant to the community. (D) The same GRNs with the top 300 nodes ranked by differential PageRank centrality between IPF and control highlighted in red. Node sizes correspond to PageRank centralities. (E) Selected results from GO gene set enrichment of the top 300 differential PageRank nodes between IPF and controls, with all nodes used as a reference. TGF-β, transforming growth factor–β; BMP, bone morphogenetic protein.