Bio-medical Ontologies Maintenance and Change Management

Protein Data Integration Problem 59
is difficult: The different database projects may use different terms to refer to the
same concept and the same terms to refer to different concepts. Furthermore, these
terms are typically not formally linked with each other in any way. GO seeks to
reveal these underlying biological functionalities by providing a structured controlled
vocabulary that can be used to describe gene products, and is shared between
biological databases. This facilitates querying for gene products that share
biologically meaningful attributes, whether from separate databases or within the
same database.
Association between ontology nodes and proteins, namely, protein annotation
through gene ontology, is an integral application of GO. To efficiently annotate
proteins, the GO Consortium developed a software platform, the GO Engine,
which combines rigorous sequence homology comparison with text information
analysis. During evolution, many new genes arose through mutation, duplication,
and recombination of the ancestral genes. When one species evolved into another,
the majority of orthologs retained very high levels of homology. The high
sequence similarity between orthologs forms one of the foundations of the GO
Engine.
Text information related to individual genes or proteins is immersed in the vast
ocean of biomedical literature. Manual review of the literature to annotate proteins
presents a daunting task. Several recent papers described the development of various
methods for the automatic extraction of text information (Jenssen et al., 2001,
Li et al., 2000). However, the direct applications of these approaches in GO annotation
have been minimal. A simple correlation of text information with specific
GO nodes in the training data predicts GO association for unannotated proteins.
The GO Engine combines homology information, a unique protein-clustering procedure,
and text information analysis to create the best possible annotations.
Recently, Protein Data Bank (PDB) also released versions of the PDB Exchange
Dictionary and the PDB archival files in XML format collectively named
PDBML (Westbrook et al., 2005). The representation of PDB data in XML builds
from content of PDB Exchange Dictionary, both for assignment of data item
names and defining data organization. PDB Exchange and XML Representations
use the same logical data organization. A side effect of maintaining a logical correspondence
with PDB Exchange representation is that PDBML lacks the hierarchical
structure characteristic of XML data. A directed acyclic graph (DAG) based
ontology induction tool (Mani et al., 2004) is also used to constructs a protein ontology
including protein names found in MEDLINE abstracts and in UNIPROT. It
is a typical example of text mining the literature and the data sources. It can’t be
classified as protein ontology as it represents only the relationship between protein
literatures and does not formalize knowledge about the protein synthesis process.
Recently, the Genomes to Life Initiative was introduced (Frazier et al., 2003a,
Frazier et al., 2003b) close to completion of Human Genome Project (HGP) which
finished in April 2003 (Collins et al., 2003). Lessons learnt from HGP will guide
ongoing management and coordination of GTL. GTL states an objective: “To correlate
information about multiprotein machines with data in major protein databases
to better understand sequence, structure and function of protein machines.”
This objective can be achieved to some extent only by creating a Generic Protein