Making hypotheses precise: applying the physical method and linear modeling to evolutionary genomics

Abstract

Hypotheses in evolutionary genomics are commonly tested by qualitative comparisons among variables. However, the number of possible outcomes for such predictions is usually few, which can potentially trivialize or even mislead conclusions. Therefore, I compared the utility of two alternative approaches to testing evolutionary genomic hypotheses. The first is the “physical method”, the quantitative approach often used in physical sciences, which I employed to mathematically model the process of gene movement between chromosomes. Using this model, I rejected the hypothesis that sexual conflict has caused biased movement of oxidative phosphorylation (OXPHOS) genes off the X chromosome in mammals. In contrast, the qualitative approach of testing for an under-representation of OXPHOS genes on the X did not yield a clear conclusion. The second alternative is to estimate effect sizes of explanatory variables using linear modelling. I tested the utility of this approach by assessing a prior study in which p-values alone were used to conclude that gene expression level is the only important factor affecting evolutionary constraint on OXPHOS genes in animals. I show that, by considering how effect sizes are estimated, it is clear that at least one additional factor must affect the evolutionary constraint on OXPHOS genes, or that the original data violated at least one assumption of linear modeling. Together, these results suggest quantitative approaches offer a more precise description of hypotheses in evolutionary genomics, which allows for more robust conclusions.