Fig. 3

Inference accuracy of A fitness effects and B establishment times (FitMut2). Comparison of inferred s and \(\tau\) with simulation. Each panel in the \(4\times 4\) array corresponds to one simulation (Methods section). Each point is an adaptive mutation that established in the simulation and was identified by FitMut2. Points in (A) are colored by their true occurrence time \(t^*\), while points in (B) are colored by their true fitness \(s^*\). Negative \(t^*\) indicates adaptive mutations that occurred during pregrowth. \(\epsilon _{\text {rel}}\) in (A) is defined as \(\frac{|s^* - \hat{s}|}{s^*}\) and \(\epsilon _{\text {abs}}\) in (B) is defined as \(|t^* - \hat{\tau }|\). \(\rho _p\) is the Pearson correlation coefficient. The 5th column in (A) shows the comparison between \(\mu (s)\) and the inferred DFE (estimated as in Supplement S8). In Fig. S2 we show the same data for FitMut1