View Program - 2022 Population, Evolutionary, and Quantitative Genetics Conference

Forward-in-time simulations are a valuable tool in population genetics that offer a chance to capture complex evolutionary dynamics but are more computationally expensive than commonly used coalescent simulations. Scaling of such forward-in-time simulations is a popular method of alleviating this computational burden by scaling the number of simulated individuals and the duration of the simulated time period (in generations), while upscaling other parameters such as the mutation rate by the same value. This rescaling approach is based on assumptions that may be violated in commonly simulated scenarios (e.g. those with natural selection), potentially biasing results. However, there have been few studies on the effects of such scaling on the accuracy of results. In this study, we carry out both unscaled and scaled simulations under a variety of demographic and selective scenarios using various scaling factors, and then use statistical and machine learning tools to investigate how key simulation results and summary statistics diverge from their unscaled counterparts. Our results indicate that some properties of simulations, such as the site frequency spectrum, are relatively robust to modest rescaling, while others such as the distribution of mutation loss times can be biased substantially.

To Scale or Not to Scale: The Influence of Scaling on Forward-in-time Population Genetics Simulations