In Silico Hypothesis Discovery - Translational Informatics: Realizing the Promise of Knowledge-Driven Healthcare - page 130

for an exploration of ensuing in silico hypothesis discovery methods, as will be
introduced in Sect. 8.3 . Additionally, Payne et al. [ 29 ] provide a more comprehen-
sive review of the theories, frameworks, and methods that make up the biomedical
KE domain.

8.3 Design and Use of Intelligent Agents for In Silico

Hypothesis Generation

While there exist a broad variety of methods that can be used for the purposes
of in silico hypothesis discovery, spanning a spectrum from machine learning
and data mining to iterative human-computer interaction in order to discovery
high level patterns within complex data sets, for the purposes of this chapter, we
will focus on a specifi c and exemplar type of methodology known as knowledge
discovery in databases (KDD). This specifi c method has been selected in order
to highlight the generalizable features of a much broad class of knowledge-based
software and intelligent agents that can be used for in silico hypothesis genera-
tion. At a high level KDD is concerned with the utilization of intelligent agents,
which are software applications that are designed to replicate human problem
solving through the leverage of conceptual knowledge collections as an integral
part of their architecture and function. In KDD, intelligent agents are used spe-
cifi cally to derive knowledge from the contents of databases, including database
metadata. The use of domain-specifi c conceptual knowledge collections, such as
ontologies, is central to the KDD induction process since commonly used data-
base modeling approaches do not incorporate semantic knowledge corresponding
to the database contents. This overall approach is the basis for a specifi c KDD
methodology known as constructive induction (CI). In CI, data elements defi ned
by a database schema are mapped to concepts defi ned by one or more ontologies
or equivalent conceptual knowledge collections. Subsequently, the relationships
included in the mapped ontologies are used to induce semantically meaningful
relationships between the mapped data elements. The induction process gener-
ates what are known as “facts” concerning the contents of the database, which are
defi ned in terms of data elements and semantic relationships that signifi cantly link
those elements together (Fig. 8.6 ).
These “facts” (which are a type of conceptual knowledge) can then be used
to support higher level reasoning about the data defi ned by the targeted database
schema. It is important to note that such “facts” can exploit the transitive closure
principles associated with the graph-like representation of most ontologies, and
therefore may include intermediate concepts that do not map to a database element
but serve to create a semantically related concept triplet or high-order relationship
that begins and terminates with concepts that do map to database elements.
The implementation of an intelligent agent that utilizes the preceding CI method-
ology often follows the multi-step process illustrated in Fig. 8.7 (which each phase
numbered to refl ect the following description) and outlined below: