Fig. 3

Statistical models of epistasis. a The fitness effects of three example insertion mutations on background strains obtained from a yeast cross (data from [44]). These examples illustrate a mutation that shows no detectable epistatic interactions (left), a mutation that has a specific interaction with one locus in the background (middle) and a mutation that shows a negative linear trend (right), which arises if a mutation has many idiosyncratic interactions with background loci. b Theory predicts that the slope of the linear global epistasis trend depends on the number and magnitude of interactions the mutation has with background loci. The variability in slopes across mutations can potentially be explained by their involvement in a variable number of pathways that contribute to fitness. c An empirical fitness landscape with ten mutations (85% complete) shows a linear global epistasis trend for mutation PMA1. This trend emerges as terms of increasing orders of epistasis are added to the model. Data from [53]