The rights and wrongs of rescaling in population genetics simulations
Computer simulations of complex population genetic models are an essential tool for making sense of the large-scale datasets of multiple genome sequences from a single species that are becoming increasingly available. A widely used approach for reducing computing time is to simulate populations that are much smaller than the natural populations that they are intended to represent, by using parameters such as selection coefficients and mutation rates whose products with the population size correspond to those of the natural populations. This approach has come to be known as rescaling, and is justified by the theory of the genetics of finite populations. Recently, however, there have been criticisms of this practice, which have brought to light situations in which it can lead to erroneous conclusions. This paper reviews the theoretical basis for rescaling, and relates it to current practice in population genetics simulations. It shows that some population genetic statistics are scaleable while others are not. Additionally, it shows that there are likely to be problems with rescaling when simulating large chromosomal regions, due to the non-linear relation between the physical distance between a pair of separate nucleotide sites and the frequency of recombination between them. Other difficulties with rescaling can arise in connection with simulations of selection on complex traits, and with populations that reproduce partly by self-fertilization or asexual reproduction. A number of recommendations are made for good practice in relation to rescaling.