Figure 4. Robust ranking of drug pairs.

(A) Four hypothetical JDFEs, ranked by their $c$ statistic. For all four JDFEs, the mean and the standard deviation in the home environment are $-1\times {10}^{-3}$ and ${\displaystyle 0.01}$, respectively. The mean and the standard deviation in the non-home environment are $1\times {10}^{-4}$ and $5.1\times {10}^{-3}$ (rank 1), $2.6\times {10}^{-3}$ and $7.5\times {10}^{-3}$ (rank 2), $5.1\times {10}^{-3}$ and $7.5\times {10}^{-3}$ (rank 3), $7.5\times {10}^{-3}$ and ${\displaystyle 0.01}$ (rank 4). Correlation coefficient for all four JDFEs is $-0.9$. (B) Collateral resistance risk over time, measured as the fraction of populations with positive mean fitness in the non-home environment. These fractions are estimated from 1000 replicate Wright-Fisher simulation runs with $N={10}^4$, $U={10}^{-4}$ ($N{U}_b=0.46$). Colors correspond to the JDFEs in panel A. Numbers indicate the -rank of each JDFE. (C) A priori $c$-rank ($x$-axis) versus the a posteriori rank ($y$-axis) based on the risk of collateral resistance observed in simulations, for all 125 Gaussian JDFEs and all $N{U}_b$ values shown in Figure 3. Gray dashed line is the identity line. $R^2$ values are ${\displaystyle 0.94,0.81}$ and ${\displaystyle 0.82}$ for $N{U}_b=0.46,4.6$ and ${\displaystyle 46}$, respectively. $P\S lt;{10}^{-15}$ for all regressions.

Figure 4—figure supplement 1. Ranking of AMP concentrations according to their risk of collateral resistance, based on Stiffler et al., 2015 data.

(A) Same as Figure 1—figure supplement 1 but colored by $c$-rank, as indicated. (B) Home and non-home fitness trajectories for the JDFEs shown in the corresponding panels in A. Notations are as in Figure 2. Population size $N={10}^4$, mutation rate $U={10}^{-4}$. (C) Collateral resistance risk over time estimated from 5000 replicate simulations shown in B. Colors correspond to the ranks in A.