Along with molecular data, clinical data (e.g., patient history, laboratory anal-
ysis, ultrasound parameters, etc.) have been shown to be very useful inpersonalized
medicine(PM). PM deals with identification of risk factors and adaptation of
management and treatment to individual patients.^15 Unlike traditional diagnosis
and treatment, personalized medicine hopes to offer better, data-driven diagnoses
based on genomic information and clinical data, which could lead to earlier
intervention and more efficient drug deployment and treatment.^16 PM could trans-
form the global health care industry, and eventually it could make a great impact on
human life expectancy. However, most personalized medicine approaches still lack
biologically meaningful computational frameworks suitable for joint analyses of
various data sets.^17 To meet the objectives of personalized medicine and develop
common strategies, we need to develop integration approaches that can make
effective use of all available data. Ideally, the aim of personalized medicine with
respect to diverse biological data is to allow medical scientists with limited com-
putational skills to connect multiple layers of patient information, together with
molecular data, and to obtain more informative prognoses that would enable them
to propose more optimal treatment choices.
To handle these large amounts of diverse biological and clinical data and
efficiently deal with their heterogeneity, we need computational methods. In this
chapter, we provide a survey of biological data sources, along with computational
methods for their integration. Here, the term data integration should not be confused
withdatabase integration, which aims to make data more comprehensively avail-
able.^18 Database integration focuses more on the accessibility of the data but not on
the computational methods for data exploration (effective use of many data sets to
obtain novel insights), which is the main focus of this chapter.
We focus on three widely used, most competitive machine learning approaches:
bayesian networks for data integration,kernel-based data integration, and the
recently proposedmatrix factorization-based data integration. Our aim is to pro-
vide detailed descriptions of these methods and demonstrate their applicability and
performance, with a special focus on the following biological problems:protein
function prediction, drug repurposing, disease classification, disease association
prediction and prioritization of disease genes.
We start this chapter with a description of these biological problems that can be
addressed by integrative approaches (Sect. 2 ). Next, we make a short introduction
into molecular networks and data repositories commonly used in their construction
(Sect. 3 ). We further demonstrate their role in solving the aforementioned biolog-
ical problems. We show how these networks are used to represent proteomic,
genomic, and other types ofomicsdata and how they can be effectively integrated
with other data to infer new relations between biological entities and to obtain new
(^15) Hamburg and Collins ( 2010 ).
(^16) Dudley and Karczewski ( 2013 ).
(^17) Evers et al. ( 2012 ).
(^18) Gomez-Cabrero et al. ( 2014 ).
140 V. Gligorijevic ́and N. Pržulj