Computational aspects of antibody gene families
Ron R. Hightower · 1996
Genetic algorithms (GAs) simulate biological evolution. For the sake of simplicity they do not incorporate many of the rich mechanisms found in natural genetic systems. Some of these mechanisms have only negligible effect on the process of evolution, but others are quite potent. Researchers in genetic algorithms have the appropriate background and tools to explore the characteristics of the ignored mechanisms and discover their potential benefits. In this dissertation I explore the evolution of the antibody molecule, which is affected by four mechanisms not explicitly incorporated into the typical GA. The first is multigene families--groups of genes with close linkage and overlapping phenotype function. The second is stochastic gene expression, which is related to the presence of multigene families. The third is incomplete fitness evaluation, which has been explored only in part by GA practitioners. The fourth is somatic learning. All four of these mechanisms influence the evolution of the antibody molecule. I introduce a model of the antibody multigene families and evolve the model using the genetic algorithm. I look at the effect of partial gene expression and incomplete fitness testing. I then look at the interaction between learning and evolution, treating clonal selection as a learning algorithm. These mechanisms are shown to improve the performance of the genetic algorithm.