OBSERVED DYNAMICS OF LARGE SCALE PARALLEL EVOLUTIONARY ALGORITHMS WITH IMPLICATIONS FOR PROTEIN ENGINEERING

The 'bi' and 'higher modal features' are aspects of Evolutionary Algorithm (EA) behaviors that are revealed, for a wide range of conditions, when extensive parametric studies are done to explore convergence time over a wide range of mutation rates. The bimodal feature indicates optimal mutation rates in terms of convergence time, which often correspond to optimal mutation rates in terms of final solution quality. The significance of the bimodal feature lies in parameter setting issues, and it is of interest to see how it varies with parameters and EA designs. Previous work shows that it appears in a wide range of conditions, but attenuates (the local optimum in convergence time becomes less apparent) with larger population sizes and low selection pressure. This chapter extends exploration of the bimodal feature into EAs with much larger population sizes, and shows that under sufficiently high selection pressure it 'returns'. It is interesting to note that these observations apply directly in the emerging field of 'Directed Evolution' for novel bio-molecules, in which large parallel populations undergo evolutionary search, with solution quality and number of generations being vital to optimise. This has potentially highly significant consequences for setting of mutation rates in Directed Evolution and high selection pressure large-scale parallel EAs in general.