Fig. 7.

Performance improves as the number of combined replicates increases. Spearman’s rank correlation coefficients between inferred and true selection coefficients averaged over 200 simulations with same setup as shown in Fig. 1a are shown as we increase the number of combined replicates under different limited sampling effects. The performance of regularized methods (linear-cov and nonlinear-corr) generally converges with 5–10 replicates. More replicates are needed when the length of data is shorter and when the sampling time interval is larger.