Publication Date


Date of Final Oral Examination (Defense)


Type of Culminating Activity


Degree Title

Master of Science in Computer Science


Computer Science

Major Advisor

Timothy Andersen, Ph.D


Jeffrey W. Habig, Ph.D


Elena A. Sherman, Ph.D


The ability of science to produce experimental data greatly surpasses our current ability to effectively visualize, conceptualize, and integrate the vast volumes of available data into a unified understanding of how complex biological systems work. This inability is a hindrance to scientific progress, and is particularly daunting when one considers multidimensional and shape-based observations as in the field of regenerative biology. For example, for at least the last 200 years, scientists have been interested in the exceptional ability of Planaria to regenerate lost tissues from damage, and there is a large amount of experimental data available on this organism. However, until recently, none of these experiments had been collected into a single database. To this end, a repository (PlanformDB) has been created that includes formal descriptions of planaria experiments, including morphological descriptions of the worms using a graph formalism. PlanformDB opens the door to automated, formal approaches for analyzing and understanding the large amount of available experimental data for planaria.

This work seeks to automate the search for models of planaria regeneration against the Planform database with experiments. Regeneration models not only help the understanding of how planarians maintain their shape based on the experiments observed up to today, but also provide a tool to predict the outcomes of future experiments. An automated model discovery framework was setup to simulate the experiments described in PlanformDB using an agent-based modeling platform combined with evolutionary search to identify plausible mechanisms for the biological behavior. The automation has been achieved through the linking of the simulation platform to PlanformDB and development of fitness metrics that enable the evolutionary search.

The proposed fitness metrics were developed, implemented and then evaluated by assessing their fitness landscapes. A fitness landscape represents the range of possible fitness values that can be assigned to various models. In this work, the roughness, flatness and the presence of local maxima in the fitness landscapes were evaluated for the proposed fitness functions. To further test the utility of the proposed fitness functions, a simple evolutionary search was performed.