FX174 Attenuation by Whole Genome Codon Deoptimization

James T. Van Leuven; Martina M. Ederer; Katelyn Burleigh; LuAnn Scott; Randall A. Hughes; Vlad Codrea; Andrew D. Ellington; Holly Wichman; Craig Miller

We analyzed the network fitness data from genes A, F, and H using the Stickbreaker R package (88) and functions therein. This package fits such data to the additive, multiplicative, and stickbreaking models. While the additive and multiplicative models assume a mutation (a recoded block in this context) changes background fitness by a difference or a factor (respectively), the stickbreaking model assumes a mutation's effect is scaled by the distance between the background and a fitness boundary. For fitting the stickbreaking model, we could not obtain reasonable estimates for the fitness boundary from the data (beneficial mutations are much more useful for estimating the boundary than deleterious ones). Instead, we assumed a fitness boundary of 24.5 dbl/hr (wildtype has fitness of 20.5); using a larger fitness boundary simply makes the sticking model more and more like the additive model. Relative fit (posterior probabilities) was calculated following the methods in (88) . To estimate the absolute goodness of fit, we used parametric bootstrap. Specifically, for each gene and each model, we extracted the observed effect of each block on each background it appeared on. For each recoded fragment, we then regressed the background's fitness against the fitness effect by fitting a simple linear model and obtained a p-value associated with a slope of zero (illustrated for gene A in fig 5) . When a model is correct, the slope of this line is expected to be zero. For a given gene and model, we take the sum of the logs of the p-values, P obs , as a summary statistic.