Mapping the drivers of within-host pathogen evolution using massive data sets

Palmer, Duncan S.; Turner, Isaac; Fidler, Sarah; Frater, John; Goedhals, Dominique; Goulder, Philip; Huang, Kuan-Hsiang Gary; Oxenius, Annette; Phillips, Rodney; Shapiro, Roger; Vuuren, Cloete van; McLean, Angela R.; McVean, Gil

These findings demonstrate the ability of our method to accurately identify many sites known to be under HLA associated selection, and highlight the nuanced effects of many putative escape variants. N (0, 4) . µ was set at 40, and κ was set at 7.60. The recombination probability between sites, r i , was set at a constant across the region, which we varied between collections of runs (r ∈ {0, 0.01, 0.05}). Figure 5 : Simulation Study 1: comparison of estimates obtained using the closest 100 sequences by Hamming distance to the sequences actually copied from in the simulation, r = 0.05. Averages are taken over 100 independent MCMC runs on independent simulated data with the same underlying parameters. Plotting for the first three rows is as in Figure 3 . The first row shows inference results for HLA 3 associated selection and reversion using sequences copied from in the restriction of D B , the second row shows the corresponding inference results when restricting D B is restricted using the closest 100 sequences by Hamming distance. The third row displays the coverage using the different restrictions. The final two rows display recombination estimates, and an example of inference of ω across the region for the restriction using the sequences copied from and the closest 100 by Hamming distance respectively. Source data are provided as a Source Data file. 1) . For Fryer Approx this is because we stop our threshold for estimated rates at 0. For PhyloD and PhyloD OR, it is because many sites will not be included in leaf distribution or logistic regression respectively. Source data are provided as a Source Data file. (1, 1) . For Fryer Approx this is because we stop our threshold for estimated rates at 0. For PhyloD and PhyloD OR, it is because many sites will not be included in leaf distribution or logistic regression respectively. Pink to red lines and dashed lines show the ROC curves for our approach, using different D B . From light to dark, we sampled 100%, 10% and 1% of D B from each independent run in Simulation Study 2 and estimated HLA-associated selection. Dashed lines of the same colour also include D in D B using a leave one out approach. The black ROC curve shows the results of considering only D using a leave one out approach. Source data are provided as a Source Data file. Grey and white bands display the 5% -95% and 25% -75% credible intervals. In the third row, we display the coverage for the 95% and 50% credible intervals in blue and black respectively. The solid lines are using the gold standard reference set, and the dotted lines are using reference data from all public databases. In each of the fourth rows we display the average RMSE at each site for different D B as defined in the key. Source data are provided as a Source Data file. Grey and white bands display the 5% -95% and 25% -75% credible intervals. In the third row, we display the coverage for the 95% and 50% credible intervals in blue and black respectively. The solid lines are using the gold standard reference set, and the dotted lines are using reference data from all public databases. In each of the fourth rows we display the average RMSE at each site for different D B as defined in the key. Source data are provided as a Source Data file. Grey and white bands display the 5% -95% and 25% -75% credible intervals. In the third row, we display the coverage for the 95% and 50% credible intervals in blue and black respectively. The solid lines are using the gold stan