Figure 2

Program sizes. We show how many TFBSs are generated by each transcriptional program in our model and relate this to the number of target genes and TFs associated with each program. On the left, the number of binding sites that our model predicts are explained by each program is shown on a log-scale. A by-product of our algorithm is that the programs are sorted by the number of TFBSs they are responsible for. The most frequently used transcriptional program accounted for almost 30,000 and 15.000 binding sites respectively and the smallest just for a handful. The largest programs are composed predominantly of common transcription factors and in general the smaller programs explain occurrences of rarer transcription factors. The right is a scatter plot of the programs showing the number of TFs against target genes. The area of each scatter point is proportional to the number of binding sites it is responsible for. Note the first two programs do not have any genes or targets associated with them, their distributions over TFs are very similar to the genomic distribution and they are ubiquitous.