To understand the cellular functions of genes requires investigating a variety of biological data, including experimental data, annotation from online databases and literatures, information about cellular interactions, and domain knowledge from biologists. These requirements demand a flexible and powerful biological data management system. MetNetDB is the biological database component of the MetNet platform (http://metnetdb.org/), a software platform for Arabidopsis system biology. This work describes a labeled graph model that addresses the challenges associated with biological network databases, and discusses the implementation of this model in MetNetDB.

MetNetDB integrates most recent data from various sources, including biological networks, gene annotation, metabolite information, and protein localization data. The integration contains four steps: data model transformation and integration; semantic mapping; data conversion and integration; and conflict resolution. MetNetDB is established as a labeled graph model. The graph structure supports network data storage and application of graph analysis algorithm. The node and edge labels have the same extension capability as object data model. In addition, rules are used to guarantee the biological network data integrity; operations are defined for graph edit and comparison.

To facilitate the integration of network data, which is often inaccurate or incomplete, a subgraph extraction algorithm is designed for MetNetDB. This algorithm allows subgraph querying based on user-specified biomolecules. Both exact matching and approximate matching with biomolecules in networks are supported. The similarity among biomolecules is inferred from expression patterns, gene ontology, chemical ontology, and protein-gene relationships. Combined with the implementation of Messmer's approximate subgraph isomorphism algorithm, MetNetDB supports exact and approximate graph matching.

Based on the MetNetDB labeled graph model and the graph matching algorithms, the MetNetDB curator tool is built with several innovative features, including active biological rule checking during network curation, tracking data change history, and a biologist-friendly visual graph query system.