Fig. 3.

Inference accuracy of the fitness. Comparison of the true fitness in simulation and the fitness inferred by Fit-Seq2.0 (red) and Fit-Seq (blue) for different sequencing read depths (columns) and distributions of fitness (rows). Each panel in the $3\times 3$ array corresponds to one simulation (Section Simulation). Each point corresponds to a lineage in the simulation. ${\rho }_{\mathrm{p}}$ is the Pearson correlation coefficient. ${ϵ}_{\mathrm{abs}}$ is the average absolute error, which is defined as $\mid s^{\ast }-\widehat{s}\mid$ for each lineage. The 4th column in shows comparison between the true distribution of the fitness f(s) in simulation (gray) and the inferred (blue for Fit-Seq and red for Fit-Seq2.0). Percentage is the fraction of lineages with more accurate estimation for the fitness using Fit-Seq2.0. Estimates generated with Fit-Seq2.0 have a higher Pearson correlation coefficient and lower absolute error when compared with Fit-Seq. The distribution of fitness effects estimated by Fit-Seq2.0 also closely matches the true distribution in the simulation