Multiscale genotype–phenotype mapping reveals the molecular scale as optimal for describing evolution

Predicting evolution remains a fundamental challenge in biology. If we were to re-run an evolutionary process, how likely would it be to reach the same outcome? How does the answer change depending on the biological scale? We explore these questions by analyzing a multiscale genotype-phenotype map of protein evolution through the lens of information theory. Our model explicitly tracks how mutation effects propagate across biological scales: a genetic scale of amino acid sequences, a molecular scale of biophysical traits (protein folding and binding free energies), an organismic scale of protein abundance, and a metabolic scale of protein flux.

We quantify evolutionary repeatability using Shannon entropy, finding that entropy decreases at larger scales. This indicates higher repeatability at more coarse-grained levels. To assess the significance of repeatability at each scale, we compare our results to a neutral baseline—a model where dynamics are governed solely by genetic drift. We compute the Kullback–Leibler (KL) divergence between evolutionary outcomes and the neutral expectation. Our framework allows us to identify an optimal scale of description, i.e., a level that maximizes information generated by selection (high KL divergence). Our results indicate that information gain is maximized at the molecular scale. We discuss the biological implications of this result for understanding evolutionary repeatability.