Summarizing Populations: Characterizing the Effects of Sampling in Computational Evolutionary Replay Experiments

Nikolai Escondo, Austin James Ferguson

When we sample an evolving population, how well do we capture its long-term evolutionary potential? This question underlies the validity of analytical replay experiments, which restart evolution from multiple points in a population’s history to measure how long-term potential changed over time. Analytical replay experiments are becoming increasingly popular in both wet-lab and computational evolution studies. However, the population sampling method is still often picked via necessity, and no direct comparisons between methods exist. Here, we use computational evolution on NK landscapes to test the effects of population sampling techniques on genetic potentiation. We analyze four techniques: full population snapshots, random sampling, cloning the most abundant genotype, and tracing the dominant lineage. We find that cloning the most abundant genotype consistently results in lower potentiation than full snapshots, while tracing the dominant lineage consistently results in higher potentiation. Random sampling falls somewhere in the middle, where the sampling rate controls the variation across replays. We end by analyzing how an increase in mutation rate can counterintuitively stabilize the impacts of sampling on potentiation. This work leverages the speed of computational evolution to both provide insights into the results of previous replay experiments and to illuminate design implications for future experiments. While many methodological questions about replay experiments remain, this work demonstrates that computational studies can provide insights via experiments that are intractable in wet-lab settings.