similarity and infer new drugs that can perform the same action (have the same
target) as the drugs in the group to which they belong. In addition to chemical
similarity, in the drug-based approach, new drugs can also be inferred based on
their molecular profiles produced by their action on a biological system. Usually,
experimental technique, such as gene expression microarray, provides a mea-
surement and representation of such molecular activity for each drug.^26- Disease-basedstrategies infer new drugs from the symptomatic and pathological
viewpoint. These approaches group similar diseases (with similar molecular
pathology, significant number of shared therapies, or similar side effects) and
infer novel drug-disease associations by expanding known associations between
that drug and some members of the group to the rest of the group. For a detailed
description of these approaches and their limitations, see Dudleyet al.( 2011 ).
Among all computational approaches for drug repurposing, machine learning-
based approaches are the most promising, providing highly reliable predictions.
The basis of these approaches is the construction of drug-drug and target-target
similarity matrices by applying the abovementioned strategies. These two matrices
are further used as input to many machine learning methods by which new drug-
disease associations are inferred.^27
In addition to machine learning methods, network-based methods are recently
gaining in popularity due to rapid accumulation of high-throughput data. These
methods use drug-target interaction (DTI) data for drug repurposing. They create a
target-target topological similarity matrix by computing the distance between
targets across the PPI network. Drug repurposing is achieved by inferring new
drug-target associations. Several network-based methods for drug repurposing have
been developed (see Wuet al.( 2013 ) for a recent review on network-based
approaches in drug repurposing).
However, despite the great accessibility of data, very few studies combine data
on drugs, diseases, and proteins into a single framework to predict new drug
candidates for repurposing.^28
2.3 Disease Classification and Disease-Disease Association
Prediction
Current disease classification and knowledge about disease associations is based on
pathological analysis and clinical symptoms. Such classification has served clini-
cians and diagnosticians for a long time. Disease Ontology (DO) database is an
example providing a comprehensive disease classification and unique annotation of
(^26) Hurle et al. ( 2013 ).
(^27) Ding et al. ( 2013 ) and Wang et al. ( 2013 ).
(^28) Wang et al. ( 2013 ), Yamanishi et al. ( 2008 ), Napolitano et al. ( 2013 ), and Huang et al. ( 2013 ).
Computational Methods for Integration of Biological Data 143