Making the Complex Comprehensible: Approximate Bayesian Computation and Behavioural Dynamics

University Of Southern California, Los Angeles CA

Investigators

Abstract

In this project the investigator and his colleagues focus on the need to develop mathematical approaches that can successfully analyze data of very high dimension, and make intelligible large numbers of higher-order interactions, focusing on an application involving a study of the genetic and environmental determinants of behavior in the model organism Drosophila melanogaster. The dearth of tractable methods of analysis for such datasets is a major bottleneck in the collection and analysis of molecular variation data. The focus is on a method that has arisen in response to the size of modern genetic variation datasets: approximate Bayesian computation [ABC]. The use of ABC has grown rapidly over the last decade, but is generally conducted in a relatively ad-hoc manner in that key choices are made in a subjective way. Aiming to put the method onto firmer foundations, they develop approaches that systematize some of the key choices of an ABC analysis. For example, they suggest that improvements are achieved by introducing a weighting scheme for summary statistics, and by systematically choosing those weights, along with other features of the analysis, as well as automatically constructing efficient low-dimensional summaries of the data. Such approaches are much needed in the ABC world. They then apply these methods to a system that represents a perfect test-bed: the parameterization of agent-based models of fly group behavior. This application is to an experimental framework considerably more complex than most current behavioral studies. We are in the midst of the Information Age, arguably the third great revolution experienced by humankind. In the agricultural revolution, a radical change in the way in which food was obtained led to great increases in its production rate and consequent great increases in population levels. In the industrial revolution, the wide-scale introduction of industrial machinery led to vastly increased industrial output and efficiency. Both of these prior revolutions have led to exponential increase in the rate at which their products (food and goods, respectively) were produced, and this eventually led to significant improvements in quality of life for the human species. The Information Age is also resulting in an exponential increase in the rate at which its product, information, is gathered, and holds forth the same hope of great improvements in our quality of life. However, the vast amounts of information now being generated can only be fully utilized once the ability to successfully deliver those data in understandable forms - i.e., analysis machinery - is developed. For this reason the investigators here develop systematic mathematical approaches that can successfully analyze datasets of extremely high dimension and thereby make intelligible the large numbers of higher order interactions that are likely to exist within such data. The approach the investigators develop here helps transform many aspects of analyses of behavior and well-being.