In this work, we explore the utility of a complex adaptive systems approach for studying epistasis, the nonlinear effects among genes that contribute to different disease outcomes. Due to the nonlinear interactions among the genes, data such as this is difficult to model using traditional epidemiological tools. Thus, we have developed EpiSwarm, a Swarm-based system for investigating complex genetic diseases. EpiSwarm uses genetic algorithms evolution on agents that function as condition-action rules to explain the data in their vicinity. These agents migrate and evolve as they cluster data in a two-dimensional world. Thus, EpiSwarm uses a genetic algorithms component to model genetic data. Although this system does not embody linkage learning explicitly, the goal of the system to identify genetic epistasis is an important process in linkage learning research. In EpiSwarm, the decision variables are biological genes, and identifying the epistasis is the core task.
This paper is an extended version of [9], extending and clarifying the explanations of the system and providing additional results.