Influenza Virus Drug Resistance: A Time-Sampled Population Genetics Perspective (original) (raw)
Figure 8
Likelihood surfaces for the contending generalized Pareto distribution.
Likelihood surfaces of the fit of a contending generalized Pareto distribution in the absence (A) or presence (B) of oseltamivir. The shape parameter κ (on the x-axis) determines the domain of attraction. The maximum likelihood estimate (indicated by a black dot) lies in the Weibull domain (κ<0_) in the absence of the drug, indicating a bounded distribution of fitness effects, whereas it lies in the Frechet domain (_κ>0) in the presence of the drug, indicating an underlying heavy-tailed distribution of fitness effects. Black lines represent 25 (A) and 5 (B) orders of log likelihood, respectively. The gray area represents inaccessible parameter space, because the lower bound of the estimated DFE cannot be lower than the maximum observed selection coefficient. We also show the distribution of the shape parameter κ for the GPD distribution (C).