In Silico Hypothesis Discovery - Translational Informatics: Realizing the Promise of Knowledge-Driven Healthcare - page 133

Fig. 8.8 Illustration of depth-based annotation and its implications for the induction of useable vs.
unusable (e.g., overly general) “facts” or hypotheses
to an ontology-anchored concept of the laboratory procedure that has that same
name, such a mapping could be used, when traversing the atomic units of infor-
mation and relationships that comprise an ontology, to assert a relationship
between “White Blood Cell Count” and the broad category of “Laboratory
Procedures”, which then in turn allows for the resolution of relationships with
every other known laboratory procedures subsumed by that concept. This
would be a factually accurate relationship to assert, but one that is functionally
useless for hypothesis discovery, as it is overly broad and general. Why is this
the case? Simply put, the concepts of “White Blood Cell Count” and
“Laboratory Procedure” are not of an equivalent level of granularity (e.g., the
former is much more specifi c than the latter). One approach that can serve as a
surrogate for concept granularity in the source ontologies employed by a
CI-based agent is the relative depth from the ontology root of those concepts
(Fig. 8.8 ). Using such measurements, we can then constrain “fact induction”
(Phase 4) to include only relationships between conceptual entities that exist at
a similar or deeper depth from the ontology root and therefore can be expected
to express useful and not overly generic hypothetical relationships. Doing so,
however, requires us to fi rst calculate the depth to the ontology root (or roots)
for every conceptual entity selected in Phase 2 of this process, usually using
the shortest such path as the preferred measurement when there exist more than
one path from the concept to the root of the source ontology or equivalent con-
ceptual knowledge construct.