Figure 4

Accuracy of preference inference on simulated data. Deep mutational scanning counts were simulated using the preferences in Figure 2A and realistic mutation and error rates that were uneven across sites and characters as in actual experiments. The simulations were done (A) without or (B) with sequencing errors quantified by control libraries. Plots show the correlation between the actual and inferred preferences as a function of the product of the sequencing depth N and the average per-site mutation rate \(\overline {\mu }\); real experiments typically have \(N\overline {\mu } \sim 1000\) to 2000 depending on the sequencing depth and gene length. Preferences are inferred using the full algorithm in dms_tools (top panels) or by simply calculating ratios of counts (bottom panels) using Equation 4 and its logical extension to include errors, both with a pseudocount of one. The dms_tools inferences are more accurate than the simple ratio estimation, with both methods converging to the actual values with increasing \(N\overline {\mu }\). Given files with the mutation counts, the plots in this figure can be generated as prefs_corr.pdf and ratio_corr.pdf with commands such as:.