Theory and practice of optimal mutation rate control in Hamming spaces of DNA sequences

We investigate the problem of optimal control of mutation by asexual self-replicating organisms represented by points in a metric space. We introduce the notion of a relatively monotonic fitness landscape and consider a generalisation of Fisher's geometric model of adaptation for such spaces. Using a Hamming space as a prime example, we derive the probability of adaptation as a function of reproduction parameters (e.g. mutation size or rate). Optimal control rules for the parameters are derived explicitly for some relatively monotonic landscapes, and then a general information-based heuristic is introduced. We then evaluate our theoretical control functions against optimal mutation functions evolved from a random population of functions using a meta genetic algorithm. Our experimental results show a close match between theory and experiment. We demonstrate this result both in artificial fitness landscapes, defined by a Hamming distance, and a natural landscape, where fitness is defined by a DNA-protein affinity. We discuss how a control of mutation rate could occur and evolve in natural organisms. We also outline future directions of this work