Extended Data Fig. 4: Annotation of programs by transcription factor binding sites, related to Fig. 4.
From: Causal modelling of gene effects from regulators to programs to traits
a) Number of cNMF components against solution stability measured by the Euclidean distance silhouette score of the clustering, and Frobenius error of the consensus solution, outputted by cNMF. b) Enrichment of transcription factor binding sites to program genes. Narrow peaks from ChIP-seq of transcription factors (TF) in K562 cells were used to calculate the enrichment (Methods). For significantly enriched TF-program pairs (FDR < 0.05), we tested the effect of knockdown of the TF on program activity and marked an asterisk if the KD also had an effect in the expected direction; that is, if the KD of an activator transcription factor decreased the program activity (p < 0.05), we marked it, and vice versa for repressor.
