Correlation between errors by noise and mutation over all genes. The frequency of errors by mutation versus noise plotted over all genes for 3 networks evolved at low noise σ = .1 > σ
. The abscissa showed the number of events for which the expression of the gene is switched on or off by noise. Computation was carried out as follows: Taking an evolved network (with highest fitness), we simulated the gene regulation network(GRN) dynamics without noise for a sufficiently long duration and obtained the original value of x(i) for each gene i. (i) The GRN dynamics were then simulated in the same manner under the noise level σ = .1 for over 105 different runs, and the number of switch events (i.e., the runs in which x(i) changes its sign) was computed. (ii) We generated 105 networks by changing 50 paths randomly chosen from the original network, and simulated these GRNs without noise and computed the number of switch events for each gene i. Step (i) gives the number of error by noise, and step (ii) gives the error by mutation, for each gene i. The relations between these 2 errors were plotted for 3 original networks with a different color for each network.