Opening the black box: A tool for evaluating the biological information of a GARP rule-set using examples from the natural ecology of Anthrax

Several recent papers have implemented the Genetic Algorithm for Rule-set Production (GARP) modeling system to develop ecological niche models for predicting the distribution of various disease agents and/or vector species. In brief, GARP is a genetic algorithm that evolves ecological niche models iteratively through a process of training and testing. The resulting models are rule sets, in the form of logic strings that describe the relationships between the point locality data (outbreak localities) and ecological parameters (e.g. satellite-derived environmental data layers) following classical definitions of the ecological niche. Rule sets allow the modeling system to develop a heterogeneous definition of a species’ distribution. However, the complexity of these rule-sets has generated criticism in the modeling community as a “weakness” of GARP. Specifically, critics argue that GARP is a “black box” and the rule-set complexity reduces the amount of biologically meaningful information associated with a prediction of a species’ spatial distribution. To address these criticisms, this paper will present a data-mining tool developed to scan the rule-set logic strings and extract the ecological ranges of the dominant rules within a rule set. This data-mining tool allows for specific evaluations of individual rules in both geographic and ecological space. It has proven useful for elucidating the ecological parameters that define the distribution of Bacillus anthracis, the causative agent of anthrax, in the contiguous United States. This paper will illustrate the use of this tool and discuss new insights into the spatial ecology and biogeography of B. anthracis.