Databases and Systems

74
The search can also be made against the KEGG pathway diagrams, which form a
much larger data set than the ortholog group tables. In this case the search has to be
made against a single reference organism, but the procedure is similar; the user
specifies a set of sequences as a query. When the EC numbers are preassigned to
enzyme genes in the genome, the user can match the set of EC numbers against the
KEGG reference metabolic pathways (not the organism-specific pathways). The
matched enzymes are marked by color, so that the connectivity and completeness of
the marked enzymes can be used to assess the correctness of functional assignment
(EC number assignment) in the gene catalog. The existence of a missing element
implies either the gene function assignment is wrong or the biochemical knowledge
of reaction pathways is incomplete [7].
Pathway reconstruction from binary relations
The prediction above is a homology modeling based on comparison against the welldefined
reference. Perhaps, the most challenging task in KEGG is to make
predictions even when the reference is missing or incomplete. In the case of the
metabolic pathways, if the reconstructed pathway contains a missing element and it is
not due to an error in the EC number assignment, then this implies that the reference
knowledge is incomplete; an alternative reaction pathway exists or an alternative
enzyme with wider specificity takes the place [7]. To investigate this possibility
KEGG provides a tool to compute from a given list of enzymes all possible pathways
between two compounds and allowing changes in specificity. In our representation,
because a list of enzymes is equivalent to a list of substrate-product binary relations,
the procedure involves deduction from binary relations, which is like combining
multiple links in LinkDB. Possible changes of substrate specificity can be
incorporated by considering the hierarchy of EC numbers; namely, allowing a group
of enzymes to be incorporated whenever any member of the group is identified in the
genome, which effectively increases the number of substrate-product relations. Figure
4 shows the procedure of computing chemical reaction paths in terms of the
deductive database.