Epistasis: just say no!

Epistasis, where the effect of a mutation depends on other mutations, has been observed in virtually every quantitative biochemical feature examined. Despite its ubiquity, epistasis is often difficult to explain mechanistically. Its interpretation is thus often limited to phenomenological descriptions of synergy or antagonism. Here we argue that epistasis quantifies the mismatch between experimental observations and a model for how mutations should sum together. Put differently: it is a fit residual. And like any fit residual, the presence of epistasis is a call to action, not an interesting observation in itself. It signals that the model is inadequate, not (necessarily) that the mutations physically interact. We propose that biochemists and biophysicists should pivot away from analyzing epistasis and into fitting mechanistic models that connect genotype to phenotype. Such models describe the effects of mutations as perturbations to underlying biochemical parameters-stability, binding affinity, catalytic rate-and combine those perturbations on mechanistically appropriate scales. Compared to a standard epistasis analysis, this approach reduces the number of free parameters, improves extrapolative power, and yields interpretable insights into structure-function relationships. We illustrate how this framework works in practice with studies covering a broad range of molecular systems and biochemically relevant scales. By reframing epistasis as a starting point rather than as a final observation, biochemists are uniquely positioned to build predictive models of mutational effects grounded in the physical principles governing biomolecular function.