Fig. 4

Fitting quantitative models with sort-seq data. Additive model parameters inferred from sort-seq simulations with different library designs. a The parameters of an additive model for the activity of a regulatory RNA [9] is inferred using either a likelihood-based or an information-based approach (ρ=0.99). Three values are outside the pane are poorly constrained by the information-based approach. b Fluorescence distributions in a targeted library (black) and random libraries characterized by different mutation rates (red). Here, simulated mutations contribute independently to activity, and the robustness is R=10. c Relative error (RMS error divided by input) in estimating the mean fluorescence for variants with a single mutation using the targeted (black) or random (red) libraries. d Heatmap of error as a function of the average number of mutations per variant r _mut and R. The color scale of the heatmap is such that the error in the targeted approach is white, and shades of purple and green indicate larger and smaller error, respectively. e–f Effect of interactions between mutations on estimates of single-mutation variants by an inferred additive model. Relative bias (circles) and standard deviation (bars) are plotted against the interaction power S. The dotted line indicates error of the targeted approach. g Heatmap of error as in d, plotted as a function of r _mut and S, for sequences with fixed R=10