network. For network-based approaches for protein function prediction we refer
readers to a review paper by Sharanet al.( 2007 ).
Here, we focus on integrative approaches for protein function prediction using
molecular interaction networks, ontology data, and phenotype information. Each of
these data sets contains valuable information about each individual protein. Their
fusion produces more detailed and reliable picture of the cellular context of each
protein. In Sect. 4 we introduce three commonly used machine learning methods
that are able to effectively combine all these data types and generate accurate
predictions of protein functions.
2.2 Drug Repurposing
Drug repurposing(also known asdrug repositioning) is a promising method with
the potential to revolutionize approaches for drug discovery in the pharmaceutical
industry. It aims to find new uses for existing drugs, i.e., to find new diseases that
may be treated with existing drugs.^21 This approach provides a variety of benefits,
such as reducing the traditional timeline for making drugs available on the market
(which normally is 10–17 years^22 ), reducing risks in clinical testings because the
drug candidates already have well-known safety profiles, and also reducing the
overall cost of development of new drugs. Some successful examples of drug
repurposing include the following: (a) the use of viagra for erectile dysfunction,
while its original use was in treatment of pulmonary hypertension;^23 (b) metformin
for cancer prevention and treatment, while its original use was in treatment of type
2 diabetes;^24 (c) ropinirole (requip) and pramipexole for restless legs syndrome,
while its original use was in treatment of Parkinson’s disease.^25
Drug repurposing problem provides new opportunities for computational scien-
tists to develop automated frameworks that can utilize existing, available, system-
level biological data to generate new associations between diseases and existing
drugs. Most of the computational methods exploit thepolypharmacologicalprop-
erty of drugs, that is, that a single drug can act on multiple targets. A variety of
different classifications of computational approaches in drug repurposing have been
proposed. For instance, Dudleyet al.( 2011 ) suggested a classification intodrug-
basedanddisease-basedstrategies:
- Drug-basedstrategies predict new drugs from the chemical or pharmaceutical
properties. These approaches group drugs based on their structural and chemical
(^21) Ashburn and Thor ( 2004 ).
(^22) Hurle et al. ( 2013 ).
(^23) Wu et al. ( 2013 ).
(^24) Quinn et al. ( 2013 ).
(^25) Trenkwalder et al. ( 2004 ).
142 V. Gligorijevic ́and N. Pržulj