Fig. 2

Summarised diagnostics for annotation of the human genome by average GC proportion using oHMMed with normal emission densities. A shows the mean (black) and median (grey) log-likelihood of fully converged runs of the algorithm with different numbers of hidden states, with the dashed horizontal line marking the selected number of hidden states, which is five. The difference between the mean and median for six hidden states is the effect of autocorrelation in the traces of the estimated parameters. In B, boxplots of the posterior (i.e., inferred) mean GC proportion of the run with the five hidden states are presented. C shows the observed overall density (black) of the GC proportion superimposed on the posterior (inferred) density, with the inferred means per chosen number of states plus the 68% confidence intervals drawn in vertical lines. The final D shows the QQ-plot of the observed density vs. the posterior density (here termed the theoretical distribution). Full descriptions of the diagnostics available for oHMMed can be found in our usage recommendations on GitHub [40], and the code for this visual summary is available as an R script named “oHMMedOutputAnalyses.R” [39] on GitHub