Title: Bayes' rule in the Serengeti: using probabilistic models to reveal the network structure of food webs

Speaker: Edward Baskerville, Ph.D candidate, Ecology and Evolutionary Biology & Scientific Computing, University of Michigan

Date/Time: Monday, October 25, 2010 1:00-2:00 pm Pacific NM: 2:00-3:00 pm Mountain       

Location: 915/S101 (Sandia CA) and CSRI Building/Room 90 (Sandia NM)

Brief Abstract: Food webs, collections of all the feeding relationships in an ecosystem, have been an important tool for ecologists for more than a century. The use of food web networks as objects for quantitative study, largely begun in the 1970s, has grown dramatically with increases in computational power and the development of new mathematical tools, mirroring the broader rise of network approaches in social science and other areas of biology. Recently, ecologists have begun using probabilistic models of food-web structure in a likelihood framework in order to encode and compare hypotheses. One useful model is based on groups, where the probability of a link is determined by a link density parameter between pairs of groups, which may be high (e.g., as one would expect between a group of species that all eat the same prey, and their prey) or low (e.g., within a group of plants). This is a generalized notion of the "modules" or "compartments" in most studies of networks, which require high within-group link density.

I will discuss how we have extended this modeling approach, and the group model specifically, to a fully Bayesian context, where model parameters, including arrangements into groups (i.e., graph partitions) are represented as a probability distribution rather than a single best solution. By explicitly representing uncertainty in the model in this way, we produce richer inferential information and avoid fundamental problems with maximization approaches. For model comparisons, we use the marginal likelihood, which automatically penalizes the effective degrees of freedom in a model.

I will also discuss how these tools reveal interesting relationships between spatial structure and network structure in the food web of the Serengeti, one of the most intact terrestrial ecosystems on the planet.

CSRI POC: Ali Pinar, 925-294-4683