In Silico Hypothesis Discovery - Translational Informatics: Realizing the Promise of Knowledge-Driven Healthcare - page 125

Sources of Data

Physical
or Virtual
Integrated
Data Repository

Investigators

Community

Intelligent
Agent

EHR/PHR-derived
Phenotype

Discovery, Interaction, and
Prioritization

Bio-molecular
Instrumentation

Hypotheses

Knowledge Acquisition and
Representation

Evaluation Metrics

Research-specific
Data Capture
Instruments

Conceptual
Knowledge

Data Sets and
Metadata

Extracted
Knowledge
(Rules, Literature)

Sources of Domain
Knowledge

Fig. 8.2 Alternative, high-throughput approach to asking and answering questions regarding “big
data” resources, using in silico hypothesis discovery methods. In this model, intelligent computa-
tional agents draw upon a variety of domain knowledge collections, using formally represented
variants of those collections, in order to identify potential relationships of interest between ele-
ments or collections of elements in an integrated repository. These relationships are then presented,
along with corresponding evaluation metrics that serve to characterize their potential accuracy and
novelty, to both investigators and their teams as well as broader groups of interested community
members, who can then discover, interact with, and prioritize such hypotheses concerning data-
level interactions for subsequent investigation

contents of those data sets using methods that are not far removed from those used
around the time of the dawn of modern science [ 13 ]. This concerning juxtaposition
is the driver for an emerging body of research that seeks to couple high-throughput
data generation with new and similarly high-throughput hypothesis generation tech-
niques, which can at a high level be referred to as in silico hypothesis discovery
methods (Fig. 8.2 ).
Such high-throughput approaches to asking and answering questions corre-
sponding to “big data” resources are essential to the synthesis of novel biomedical
knowledge, such as that required to support personalized medicine paradigms. Such
precision approaches to wellness promotion and care delivery aim to improve quality,
outcomes and cost of care [ 2 , 3 , 14 - 16 ]. Acting upon this vision of high-throughput
in silico hypothesis discovery requires:

1. An understanding of the design and appropriate use of domain-specifi c concep-
tual knowledge collections;
2. The application of intelligent agents that are informed by such knowledge col-
lections and based upon formal computational methods; and
3. The evaluation of ensuing hypothesis using appropriate metrics and measures.