Finally, in order to fully specify the mutation model, we have to fix the transition-to-transversion ratio k.
For implementing the nucleotide mutation model, we define the state of each site i as a codon n = { n1 n2 n3}.
> -wrap-foot> To estimate the parameters in the background mutation model, we could fit the model in Table 1 to the sequences for which silent mutations were identified.
To fit a background mutation model, we used PAML [ 35] to fit the HKY model [ 34] to the promoters that contain the characterized binding sites for each factor.
In contrast, the deleterious mutation model we present here does not need to invoke processes to augment competition between antigenic strains because a large proportion of fitness variation in the virus population arises from differences in deleterious mutation loads.
For each mutation model, we compared observed heterozygosities and expected heterozygosities simulated 10 000 times from the allele number at equilibrium using two-tailed Wilcoxon tests, where values significantly larger than expected indicate a bottleneck, and values significantly smaller than expected indicate demographic growth.
To determine if there were any existing data regarding these Salmonella mutations in animal models we performed an extensive literature search over "fluoroquinolone", "Typhimurium", "gyrA", and certain specific mutations of interest (over 400 studies met parts of these criteria).
Nevertheless, the correlation r([ h]evol, c) is close to one for all four proteins for almost all the mutation models that we simulated.
We have adapted the approximations in their Table 2, which correspond to the infinite sites mutation model as we assume here, to the metapopulation model that we are studying, which is slightly different from the one they have used.
Our simulation results are based on a certain pathway dynamics and mutation model; however, we believe that our findings generalize, which we can better understand on a genotype-phenotype space.
