Description: Presentation at the Ecological Society of America annual meeting in Memphis, TN August 8, 2006. Part of a symposium organized by Tim Keitt and Bill Fagan: Structure and Dynamics of Ecological Networks.

We parameterize a model for predicting trophic links using previously published interaction networks and phylogenetic/taxonomic trees. Interactors in given food webs are identified where possible to scientific name at the most appropriate taxonomic level so that a tree can used to search for evidence supporting or rejecting potential network links. To maximize accuracy and recall, we examine limits on the tree search, and different costs to apply to distances from the target taxa to those found in known networks. We find that negative evidence, where relatives of two organisms occur together but are not linked trophically, must be significantly down-weighted. Also, perhaps not surprisingly, recall is difficult but not impossible for large webs whose entities are well resolved to species. We also consider variation among phyla in parameterization success. We then use this model to predict links among new lists of taxa and consider properties of these webs (e.g. connectance). Finally, we present preliminary results of experiments using intelligent agents and ontologies to integrate life history datasets towards more realistic predictions.

Type: Presentation

Authors: Cynthia Parr, Tim Finin, Joel Sachs, and Andriy Parafiynyk

Date: August 08, 2006

Tags: semantic web, ecology, modeling, sparql, food web, rdf

Format: Microsoft PowerPoint (Need a reader? Get one here)

Number of downloads: 533

Access Control: Publicly Available

Available for download as

size: 4571136 bytes

Assertions:

1. (Resource) Predicting food web connectivity: Phylogenetic scope, evidence thresholds, and intelligent agents is associated with news item (News) Parr gives invited talk at ESA 2006