6.6 Retrieval of Knowledge Representations 151
com. Figure 6.2 is an example of a query presented to SKIP. The query
in this case is “What regulates the adhesiveness of integrins at the plasma
membrane of lymphocytes, and is responsible for association of PSCDs with
membranes?” The database used by the public demonstration is the NCBI
Reference Sequences (RefSeq) database (NIH 2004a). One can further restrict
the query by specifying that it must include some concepts and that it must
exclude others.
Figure 6.2 Query screen for the SKIP retrieval system.After clicking on “Run Query”, the query text is converted to a knowledge
representation as shown in figure 6.3. In this figure the knowledge represen-
tation for the query is to the right of the query. At the bottom of the figure
are references to the two documents that match the query best. Other doc-
uments that match in other ways are shown in figure 6.4. The boxes in the
knowledge representation represent instances of the concepts shown, and
the arrows between the boxes represent relationships between the concepts.
Note that “plasma membrane” is a single concept in the UMLS so it is repre-
sented using a single box rather than two boxes joined by a relationship.
SKIP uses a high-performance indexing technology that was inspired by
biological sequence matching. As discussed in chapter 7, one can find homol-
ogous sequences by using short sequences to index the sequence database.
One then extracts short sequences from the query and matches them with
the ones in the index. For nucleotide sequence matching this can actually be
done chemically by synthesizing the query sequence as an oligonucleotide
and hybridizing it with the target DNA (in single-stranded form). The SKIP
index generalizes this technique to find knowledge representations that are
homologous to a query (Baclawski 1997a).
The matching documents are arranged by how well they match the query.