Pinney et al. 10.1073/pnas.0706339104.
Fig. 5. Probabilities of loss and gain of interaction as a function of the sum of evolutionary distances (amino acid substitutions per site) traveled by a protein pair. Probabilities were calculated by considering a set of human bZIP pairs experimentally confirmed as strongly interacting or noninteracting and constructing all possible moves in 2D sequence space that transformed one pair into another. The plotted points show the proportion of moves starting from a strongly interacting pair that result in a noninteraction (probability of loss) and the proportion of moves starting from a noninteraction that result in a strong interaction (probability of gain). Logistic functions fitted to these data also are shown.
Fig. 6. Distributions of sequence-based scores for experimentally verified strongly interacting and noninteracting human bZIP pairs.
Fig. 7. Evolutionary history of the bZIP interaction network in chordates as predicted by our benchmark LZ sequence reconstruction method. An edge is drawn between two proteins if they have >50% probability of sharing a strong interaction (see Fig. 2 legend).
Fig. 8. Evolutionary history of the bZIP interaction network in chordates as inferred by the PARS maximum-parsimony method. An edge is drawn between two proteins if they are predicted to share a strong interaction (see Fig. 2 legend).
Fig. 9. ROC evaluations of the ancestral Vertebrate (i) and Teleost (ii) networks, inferred by our probabilistic method (black line) and by a parsimony-based method (gray line) with various levels of noise applied to the input interaction scores. (a) No noise added. (b) Adding Gaussian noise with a SD of 10. (c) Adding Gaussian noise with a SD of 20. AUC, area under curve.