Figure 3

Linear dependence of domain-count changes on domain occurrence. Linear dependency of the fraction of domain-count changes on the domain-count itself. Left panel : For each genome pair i the fraction $\Delta n_c^i/\Delta n^i$ of domain-count changes that involve domains of category c is shown (vertical axis) as a function of the fraction $n_c^i/n^i$ of all domains in the genome that are associated with category c (horizontal axis) for the categories 'metabolic process' (blue), 'regulation of transcription' (red), and 'protein kinase activity' (green). Each dot corresponds to the data for one pair i of closely-related genomes. Both axes are shown on a logarithmic scale. The straight-lines show least-squares fits of the form $\log [\Delta n_c^i/\Delta n^i]={\gamma }_c\log [n_c^i/n^i]+{\delta }_c$. The fitted slopes for the three categories are γ_{prot.kin.activity} = 0.56 ± 0.46, γ_reg.transcr. = 0.95 ± 0.20, and a_met.proc. = 1.48 ± 0.31. For comparison the dotted lines show linear scaling. Right panel: A 99% posterior probability interval for the slope γ _c was estimated for all selected GO categories (Methods). The fitted slopes were ordered from high to low and are shown in the right panel from left to right with the vertical bars corresponding to the 99% posterior probability intervals for each slope γ _c . The slope γ = 1, corresponding to a linear dependency, is shown as a horizontal dotted line.