Chaos Automata for Sequence Visualization

Daniel Ashlock, Cameron McGuinness, and Wendy Ashlock
Submitted to CIBCB 2015

Abstract PDF eprint

A chaos automata is a type of side effect machine that serves as a state-conditioned version of the chaos game used to visualize DNA or other linear sequence data. This study performs a parameter study to tune an evolutionary algorithm for locating chaos automata that make relatively dense renderings of two-class DNA data. Both the number of states and the population size turn out to be relatively soft parameters, but there is benefit to tuning the mutation rate. The fitness landscape is found to be rugose and to possess a large number of optima. A reporting tool called time of last innovation is used to provide additional nuance to the traditional reporting of best fitness values. Topics for additional work are outlined, including a demonstration that chaos automata can be averaged to provide an additional avenue to search for effective visualizations. The system is tested on synthetic and biological data.