In Silico Hypothesis Discovery - Translational Informatics: Realizing the Promise of Knowledge-Driven Healthcare - page 132

Fig. 8.7 Overview of major steps, resources, and outputs associated with the design and use of a
CI-based agent

￿ Phase 2 - Subset Selection from Knowledge Collection(s) : Given that many

conceptual knowledge collections contain thousands, if not hundreds of thou-
sands, of distinct atomic entities and corresponding hierarchical or semantic
relationships, such constructs can present computational challenges, such as the
tractability, computational cost, or timeliness of computational tasks applied to
such knowledge collections, which can be addressed through a process known as
search space reduction . Effectively, once Phase 1 (Metadata to Conceptual
Knowledge Entity Mapping) is complete, we can select a subset of those concep-
tual knowledge collections that directly correspond to: (1) the atomic elements
mapped to the targeted metadata; (2) the hierarchical and/or semantic relation-
ships that serve to link those atoms together; and (3) any additional atoms neces-
sary to complete the linking paths identifi ed via [ 2 ]. This allows refi nement of
the initial knowledge collections to one that is constrained to the problem-solving
task at hand.

￿ Phase 3 - Depth-based Annotation : Once we have reduced the overall search

space (Phase 2), an additional computational challenge must be addressed,
concerned with the granularity of concepts being used for reasoning purposes.
If we extend our prior example of “White Blood Cell Count” and its mapping