PERSPECTIVE

published: 23 May 2017

doi: 10.3389/fphar.2017.00298

On the Integration of In Silico

Drug Design Methods for Drug
Repurposing
Eric March-Vila1, Luca Pinzi1, Noé Sturm1, Annachiara Tinivella1, Ola Engkvist2,
Hongming Chen2 and Giulio Rastelli1*
1
Molecular Modelling & Drug Design Lab, Department of Life Sciences, University of Modena and Reggio Emilia,
Modena, Italy, 2 Discovery Sciences, Innovative Medicines and Early Development Biotech Unit, AstraZeneca R&D
Gothenburg, Mölndal, Sweden

Drug repurposing has become an important branch of drug discovery. Several

computational approaches that help to uncover new repurposing opportunities and aid
the discovery process have been put forward, or adapted from previous applications. A
number of successful examples are now available. Overall, future developments will
greatly benefit from integration of different methods, approaches and disciplines. Steps
forward in this direction are expected to help to clarify, and therefore to rationally predict,
new drug–target, target–disease, and ultimately drug–disease associations.
Keywords: drug repurposing, drug discovery, molecular modeling, chemogenomics, structure-based drug
Edited by: design, ligand-based drug design, machine learning, transcriptomics
Yuhei Nishimura,
Mie University, Japan
Reviewed by: INTRODUCTION
Antonio Macchiarulo,
University of Perugia, Italy Drug repurposing (also known as drug repositioning) aims at identifying new uses for
Yoshito Zamami, already existing drugs (Novac, 2013). In drug discovery, drug repurposing has gained an
Tokushima University Graduate increasingly important role, because it helps to circumvent preclinical development and
School of Medical Sciences, Japan
optimization issues, hence reducing time efforts, expenses and failures typically
*Correspondence associated with the drug discovery process.
:
Over the years, biological and chemical information has been generated at an ever-increasing
Giulio Rastelli
pace, marking the entrance in the so-called “big data” era (Costa, 2014). This offers the scientific
[email protected]
community new opportunities to link drugs to diseases, although this relationship is indirect and
Specialty section: relies on complex mechanisms of action. Therefore, a better understanding of the relationships
This article was submitted to between drugs and their targets, and between targets and diseases, is a key for drug repurposing.
Experimental Pharmacology and Drug Unfortunately, we are still far from understanding the overall picture, partly due to the heterogeneity
Discovery, and incompleteness of the available data. However, computational methods o ffer valuable
a section of the journal opportunities to create such links, as it will be illustrated below.
Frontiers in Pharmacology
In this perspective, different computational methods and approaches are briefly
Received: 07 April 2017 presented, and their ability to complement and integrate each other in drug repurposing
Accepted: 10 May 2017
is discussed, which will certainly gain a foothold in the future.
Published: 23 May 2017

Citation:
March-Vila E, Pinzi L, Sturm N, COMPUTATIONAL DRUG REPURPOSING STRATEGY BASED
Tinivella A, Engkvist O, Chen H
and Rastelli G (2017) On the
ON TRANSCRIPTIONAL SIGNATURES
Integration of In Silico Drug Design
Methods for Drug Repurposing. Transcriptomic data can provide a list of over- and under-expressed genes in a biological system
Front. Pharmacol. 8:298. treated by a pharmacologically active compound. The perturbation of a biological system
doi: 10.3389/fphar.2017.00298

Frontiers in Pharmacology | www.frontiersin.org 1 May 2017 | Volume 8 | Article 298

March-Vila et al. In silico Drug Repurposing

can be measured from genome wide transcriptional (GEO)2 and integrated this network with molecular profiles
responses, and the drug induced transcriptional responses and knowledge of drugs and drug targets to infer drug
represent the signature of the compound activity on biological repositioning opportunities and suggest molecular targets
systems. These molecular transcriptional signatures can then and mechanisms underlying drug effects. Jin et al. (2012)
be compared to establish therapeutic relationships between developed a novel method to repurpose drugs for cancer
known drugs and new disease indications. therapeutics by leveraging off-target effects that may
One of the most comprehensive and systematic approaches affect important cancer cell signaling pathways. The o ff-
toward leveraging the transcriptional signature approach for drug target effects of drugs on signaling proteins were
repositioning is the Connectivity Map project (Lamb et al., 2006). identified by using a hybrid model composed of a network
The publicly funded CMap database 1 initially contained profiles of component called cancer-signaling bridges and Bayesian
164 drugs and was later expanded to 1309 FDA-approved small factor regression model.
molecules. These compounds are tested in five human cell lines,
generating over 7000 gene expression profiles in the database.
The cell perturbation profile of each drug in the reference LIGAND-BASED APPROACHES IN
collection contains, for each gene measured, a rank-based DRUG REPURPOSING
measure of the change in transcriptional activity after exposure to
the drug compound, i.e., gene signatures. These signatures form Ligand-based approaches are based on the concept that
the basis of comparing drugs mechanism of action at similar compounds tend to have similar biological properties.
transcriptional level and have been successfully applied for drug In drug repurposing, these methods have been extensively
repurposing in many examples. Chang et al. (2010) used CMap used to analyze and predict the activity of ligands for new
to identify new analgesic and antinociceptive properties of targets. Public databases of bioactive molecules, such as
phenoxybenzamine, originally an anti-hypertensive drug. PubChem, ChEMBL, and DrugBank contain information
Subsequent testing using a rat inflammatory model validated the retrieved and manually curated from literature data (Wishart
analgesic activity. In contrast with CMap gene signatures, et al., 2006; Gaulton et al., 2017; Wang et al., 2017). These
biclustering methods were applied to CMap to group coregulated databases represent a huge and ever-growing reservoir of
genes with the drugs they respond to Iskar et al. (2013). This led chemical and biological information such as binding affinity,
to the identification of vinburnine, a vasodilator, and sulconazole, cellular activity, functional and ADMET data. Recent
a topical antifungal, as interesting cell cycle blockers for cancer advances in drug repurposing include the release of
therapy. databases focused on repurposed drugs, failed drugs, their
therapeutic indications, and bioactivity data (Brown and Patel,
2017; Shameer et al., 2017).
NETWORK-BASED DRUG One advantage of applying these approaches to drug
REPURPOSING repurposing is that the number of publicly accessible
compound records (more than hundred millions provided
In recent years, network-based computational biology has only by PubChem) is far greater than the number of
attracted increasing attention. It aims at organizing the deposited protein crystal structures (as of today, less than
relationships among biological molecules in the form of networks 150,000 in the Protein Data Bank) (Berman et al., 2000;
to find newly emerged properties at a network level, and to Wang et al., 2017). On the other hand, ligand-based
investigate how cellular systems induce different biological methods obviously depend on the chemical space
phenotypes under different conditions. In the network coverage of already known molecules. Moreover, a high
pharmacology framework, a network can be depicted as a overall similarity does not necessarily guarantee activity
connected graph, where each node can represent either an on a secondary target, since local structural divergences
individual molecular entity (e.g., a drug), its biological target, a in chemical scaffolds can lead to “activity cliffs” ( Stumpfe
modifier molecule within a biological process, or a target pathway, and Bajorath, 2012). This limitation, however, will
while an edge represents either a direct or indirect interaction eventually be overcome by the increase of structural
between two connected nodes. Ultimately, both the efficacy and diversity in bioactivity databases (Hu and Bajorath, 2013).
the toxicity of a drug are a consequence of the complex interplay Recently, a 2D ligand-based similarity analysis of ChEMBL
among different cellular components. A system-scale perspective combined with support vector machine models and analysis of
is therefore needed to aid modern drug discovery, especially for 3D structural information of protein–ligand complexes, identified
complex diseases, which are known to be caused by perturbation a promising set of target combinations and associated ligands
of biological networks. within the Hsp90 interactome, which are particularly suitable for
Network-based analysis has become a widely used strategy multitarget drug design (Anighoro et al., 2015). Another ligand-
for computational drug repositioning. Hu and Agarwal (2009) based method correctly predicted 23 new drug– target
created a disease-similarity network using publicly available gene associations using the similarity ensemble approach (Keiser et
expression profiles from NCBI Gene Expression Omnibus al., 2009). Pharmacophore screening has also been a

2http://www.ncbi.nlm.nih.gov/geo/
1 http://www.broadinstitute.org/cmap

Frontiers in Pharmacology | www.frontiersin.org 2 May 2017 | Volume 8 | Article 298

March-Vila et al. In silico Drug Repurposing

valuable strategy for drug repurposing (Liu et al., 2010). In STRUCTURE-BASED APPROACHES IN
this approach, a drug can be represented as a set of DRUG REPURPOSING
pharmacophoric features which can subsequently be used
to interrogate chemical compound databases to provide It is established that the similarity principle observed for
compounds with different scaffolds. ligands applies also to proteins. Proteins with similar
Complementing different levels of ligand description structures are likely to have similar functions and to
increases the chances of identifying new repurposing recognize similar ligands. In the field of drug repurposing,
possibilities. For example, Vasudevan et al. (2012) used 3D protein comparison is used as a method to identify
shape-based descriptors to compare approved drugs with a secondary targets of an approved drug (Ehrt et al., 2016).
set of H1 receptor antagonists. Thirteen of the 23 tested From a global point of view, proteins can be compared by
drugs selectively inhibited histamine-induced calcium release sequence similarity. Protein sequences have been used to build
by acting at the H1 receptor level. Interestingly, these drugs phylogenetic trees, the most popular of which is represented by
would not have been detected with 2D similarity searching the kinome (Manning et al., 2002). In this tree, proteins of the
(Vasudevan et al., 2012). Furthermore, Mervin et al. (2015) same family are prone to have related functions and also to
demonstrated how the inclusion of inactive data improved recognize related substrates or ligands, such as for example dual
early recognition abilities in statistical prediction models. On inhibitors of epidermal growth factor receptor (EGFR) and
different grounds, predictive models built upon disease epidermal growth factor receptor B2 (ErbB2) (Zhang et al., 2004).
feature descriptors, large-scale drug–target and target– Modern methods to perform multiple-sequence alignments, such
disease associations showed performance improvements in as BLAST, are nowadays widely used and available through
predicting new drug–disease links (Iwata et al., 2015; Sawada web-servers. It is important to note that small differences
et al., 2015). In particular, it was shown that chemical localized at key positions, such as those occurring in
similarity and phenotypic similarity are complementary to correspondence of the gatekeeper residue of protein kinases or
each other, and that integrating predictions from both of other oncogenic mutations, may have a huge impact on ligand
methods is beneficial (Sawada et al., 2015). binding (Huang and Fu, 2015). Hence, local differences in
globally conserved protein sequences should be given careful
consideration. Moreover, a study based on the similarity
LIGAND-BASED CHEMOGENOMICS
ensemble approach showed that similar ligands were able to
AND MACHINE LEARNING IN DRUG bind proteins with distantly related sequences (Keiser et al.,
REPURPOSING 2007). Overall, local binding site similarities can be more
important than global similarities to determine polypharmacology
A variety of in silico approaches have been applied in ligand-
and drug repurposing (Jalencas and Mestres, 2013b; Anighoro et
based chemogenomic campaigns (Mestres et al., 2006; Bender
al., 2015).
et al., 2007; Gregori-Puigjané and Mestres, 2008). During the last
In identifying unknown targets of known ligands, sequence
years, machine learning algorithms, which span from the older
alignments perform well when proteins share a high degree
but still attractive Bayesian classifiers to the more advanced
of sequence identity, whereas local protein comparison
support vector machines, have become increasingly popular to
performs better when proteins share low sequence identity
assist the drug repositioning process (Bender et al., 2007).
(Chen et al., 2016). Detecting local similarities by comparing
Methods such as deep learning and multi-task learning have
protein binding sites has become increasingly important (Ehrt
been successfully used in chemogenomic benchmark studies
et al., 2016). Binding site identification and comparison are
(Unterthiner et al., 2014). Moreover, matrix factorization methods
commonly performed by scanning the protein surface in order
offer the opportunity to combine bioactivity data with other to identify cavities (Laurie and Jackson, 2006) and then by
information, such as disease information, in one framework calculating descriptors of different nature useful to derive a
(Zhang et al., 2014). On a different line, other techniques inspired similarity score.
by e-commerce websites have shown interesting results in It is important to note that several approaches and algorithms
identifying new drug–target associations (Alaimo et al., 2016). In for binding site comparison have been put forward, but none of
the study, the technique relies on a network-based inference them appears to be devoid of failures or limitations (Ehrt et al.,
algorithm and a drug–target bipartite graph extracted from 2016). Notwithstanding, binding site similarity has proven a
DrugBank. It was shown that the algorithm performed better in valuable tool in a number of studies. For example, a study
predicting new drug–target associations when target and drug carried out by Defranchi et al. (2010) used a binding site
similarities are considered. comparison method to predict the cross-reactivity of four protein
Given the versatility in their use and their computational kinase inhibitors with Synapsin I. These discoveries were
efficiency, machine-learning approaches will likely continue to supported by sub-micromolar affinities of the kinase inhibitors for
play a prominent role in in silico chemogenomics. Despite Synapsin I. Interestingly, binding site similarity and other
many papers have described test cases and various types of molecular modeling techniques were used in combination to
method development, there is still a lack of published success uncover new targets of the drugs entacapone and tolcapone
stories that employed ligand-based chemogenomics modeling (Kinnings et al., 2009). The study started from a large set of
in drug repurposing. similar binding sites, which was further

Frontiers in Pharmacology | www.frontiersin.org 3 May 2017 | Volume 8 | Article 298

March-Vila et al. In silico Drug Repurposing

FIGURE 1 | Connecting drugs, targets and diseases with in silico methods. Like in a network representation, the integration of different computational
methods and approaches will greatly help us advance our understanding and prediction of the complex interplay between drugs, targets, and diseases.

finalized by simulating the binding mode of entacapone networks connecting chemical fragments to chemoisosteric
and tolcapone using docking. Proteins for which ligands protein environments. These networks, complemented with
gave the best docking scores were prioritized and further target–disease associations, constitute attractive starting
experimentally validated. points for drug repurposing efforts.
It is worth mentioning that ligand binding modes, when Based on similar concepts, a method for interrogating
available, are a strong asset in the process of identifying new large data sets of proteins (as large as the PDB) with
targets. One way to model the molecular recognition is to focus highly customizable geometric patterns as searching
on target–ligand interactions. This can be achieved with various templates was recently described (Inhester et al., 2017).
methods, such as structure-based pharmacophores or interaction This method was able to identify chemoisosteric protein
fingerprints. When the structure of a protein–ligand complex is environments binding the uracil moiety of uridine
not available, one can use computational methods to predict hot diphosphate from a query built with deoxythymidine.
spots in the binding site (Hall et al., 2015). Another approach Structure-based methods are obviously dependent on the
joining ligand information to protein environments uses the availability of crystallographic structures of protein–ligand
concept of chemoisosterism (Jalencas and Mestres, 2013a). complexes. Resolution and sensitivity to atomic coordinates
Chemoisosterism can be defined as the property of two protein impact the level of details that one can use to model a binding
environments to bind the same molecular fragment, and can site. While crystallographic structures represent a static model of
shed light into the inherent cross-pharmacology between protein a protein, other pockets may appear upon conformational
targets. The degree of chemoisosterism was found to be related changes. Detecting those cryptic sites has become an emerging
to the polypharmacology of chemical fragments (Jalencas and field of research, because it may provide additional options in
Mestres, 2013b). This approach allows the creation of interaction drug repurposing. In fact, cryptic allosteric sites may be useful to
gain selectivity, explore new chemical spaces for

Frontiers in Pharmacology | www.frontiersin.org 4 May 2017 | Volume 8 | Article 298

March-Vila et al. In silico Drug Repurposing

drug design, and establish drug–target associations beyond INTEGRATING DIFFERENT APPROACHES
the more commonly explored orthosteric site. For instance, AND FUTURE DIRECTIONS
Markov models have been applied in combination with
experimental assays using a chemical probe to uncover The goal of drug repurposing is to uncover new links between
cryptic allosteric sites of TEM-1 b-lactamase (Bowman et al., drugs and diseases, most commonly via targets. As illustrated in
2015). Overall, uncovering new allosteric sites in proteins the previous sections, computational predictions followed by
may provide far more opportunities to repurpose drugs than is experimental assessment have been successfully used to
currently recognized. identify new drug repurposing possibilities. As always, each
computational method has its own field of applicability,
drawbacks and limitations. One should be aware of the fact that
MOLECULAR DOCKING none of these methods alone will be sufficiently able to disclose
(or even model) the complex interplay between drugs, targets
Molecular docking is a versatile tool used to predict the and diseases. Therefore, we are left with the possibility of using
geometry and to score the interaction of a protein in complex one or more computational approaches to “navigate” through the
with a small-molecule ligand (Kitchen et al., 2004). Therefore, wealth of available information and hopefully find “clues” solid
these methods can be used to predict if a given drug enough to justify a repurposing hypothesis worth of experimental
is potentially able to bind other targets. Docking studies investigation. The choice of the most appropriate method(s) will
have been successfully exploited in drug repurposing, as basically depend on the nature of the problem to solve and on
reported in many recent studies (Kinnings et al., 2009; Li et the type, quality, and quantity of information available on that
al., 2011; Dakshanamurthy et al., 2012). In this context, virtual problem in the literature or in public or proprietary databases.
screenings can be performed either by docking a known drug Unfortunately, information is often fragmented, and generally
into a large set of different target structures, or by docking a reflects only a single or few aspects of a much more complicated
database of approved drugs into one intended specific target. story. Future efforts should be more thoroughly directed toward
Molecular docking is in fact a convenient and fast method to disclosing hubs and links of the complex network that relates
screen large libraries of both ligands and targets, with a full drugs, targets and diseases. Integrating the huge and
range of sampling options (Kitchen et al., 2004), and is heterogeneous amount of available data (chemical, biological,
obviously restricted to studies in which a 3D structure of the structural, clinical) into a unified workflow is obviously a
target is available through crystallography, nuclear magnetic challenging task. In this respect, the integration and use of
resonance (NMR), or comparative models. It should be noted different computational methods as shown above will provide
that docking methods still have drawbacks and limitations, valuable opportunities to extend the domain of applicability of
mainly arising from the use of approximate scoring functions each method and more thoroughly exploit information coming
and imperfect binding mode placement algorithms. Often from different sources (Figure 1). Likewise, this will greatly
these problems can be overcome by post-processing docking benefit from better integration of multidisciplinary work. A
results with more accurate scoring functions and/or other network-based approach built upon these considerations will
criteria (Sgobba et al., 2012). likely provide new routes to navigate through all the potential
links between drugs and diseases, thus creating new
In the study of Li et al. (2011), docking methods have opportunities for drug repurposing and drug discovery in general.
been successfully exploited as a stand-alone method in
drug repurposing, by docking the drugs of the DrugBank
database into 35 crystal structures of MAPK14. The study
identified the chronic myeloid leukemia drug nilotinib as a AUTHOR CONTRIBUTIONS
potential anti-inflammatory drug with an in vitro IC50 of 40
nM (Li et al., 2011). All authors contributed in writing and editing the
Docking is notably well suited for either drug-based and manuscript. EM-V, LP, NS and AT contributed equally. GR
target-based drug repurposing, as reported in the results of conceived the study and coordinated the writing.
the work of Dakshanamurthy et al. (2012), where an anti-
parasitic drug was successfully tested as an anti-angiogenic
vascular endothelial growth factor receptor 2 (VEGFR2) FUNDING
inhibitor, and a new connection was discovered between
previously untargeted Cadherin-11, implied in rheumatoid The project leading to this article has received funding (for EM-V,
arthritis, and cyclooxygenase-2 (COX-2) inhibitor celecoxib. OE, HC, and GR) from the European Union’s Horizon 2020
It is important to note that docking, despite its limitations, is a research and innovation program under the Marie Skłodowska-
well-established and experimentally validated approach for Curie grant agreement No 676434, “Big Data in Chemistry”
predicting new drug–target associations. Once integrated with (“BIGCHEM”, http://bigchem.eu). The article reflects only the
ligand-based methods and other available information about authors’ view and neither the European Commission nor the
target–disease associations, it constitutes a powerful approach to Research Executive Agency are responsible for any use that may
repurpose (newly) targeted drugs for a specific disease. be made of the information it contains.

Frontiers in Pharmacology | www.frontiersin.org 5 May 2017 | Volume 8 | Article 298

March-Vila et al. In silico Drug Repurposing

REFERENCES phenotypic and molecular data. J. Chem. Inf. Model. 55, 446–459. doi:
10.1021/ ci500670q
Alaimo, S., Giugno, R., and Pulvirenti, A. (2016). Recommendation Jalencas, X., and Mestres, J. (2013a). Chemoisosterism in the proteome.
techniques for drug-target interaction prediction and drug repositioning. J. Chem. Inf. Model. 53, 279–292. doi: 10.1021/ci3002974
Methods Mol. Biol. 1415, 441–462. doi: 10.1007/978-1-4939-3572-7-23 Jalencas, X., and Mestres, J. (2013b). Identification of similar binding sites
Anighoro, A., Stumpfe, D., Heikamp, K., Beebe, K., Neckers, L. M., to detect distant polypharmacology. Mol. Inform. 32, 976–990. doi:
Bajorath, J., et al. (2015). Computational polypharmacology analysis of 10.1002/minf. 201300082
the heat shock protein 90 interactome. J. Chem. Inf. Model. 55, 676– Jin, G., Fu, C., Zhao, H., Cui, K., Chang, J., and Wong, S. T. C. (2012). A novel
686. doi: 10.1021/ ci5006959 method of transcriptional response analysis to facilitate drug repositioning for
Bender, A., Scheiber, J., Glick, M., Davies, J. W., Azzaoui, K., and Hamon, cancer therapy. Cancer Res. 72, 33–44. doi: 10.1158/0008-5472.CAN-11-2333
J. (2007). Analysis of pharmacology data and the prediction of adverse Keiser, M. J., Roth, B. L., Armbruster, B. N., Ernsberger, P., Irwin, J. J.,
drug reactions and off-target effects from chemical structure. and Shoichet, B. K. (2007). Relating protein pharmacology by ligand
ChemMedChem 2, 861–873. doi: 10.1002/cmdc.200700026 chemistry. Nat. Biotechnol. 25, 197–206. doi: 10.1038/nbt1284
Berman, H. M., Westbrook, J., Feng, Z., Gilliland, G., Bhat, T. N., Weissig, Keiser, M. J., Setola, V., Irwin, J. J., Laggner, C., Abbas, A. I., Hufeisen, S.
H., et al. (2000). The protein data bank. Nucleic Acids Res. 28, 235– J., et al. (2009). Predicting new molecular targets for known drugs.
242. doi: 10.1093/nar/ 28.1.235 Nature 462, 175–181. doi: 10.1038/nature08506
Bowman, G. R., Bolin, E. R., Hart, K. M., Maguire, B. C., and Marqusee, Kinnings, S. L., Liu, N., Buchmeier, N., Tonge, P. J., Xie, L., and Bourne, P. E. (2009).
S. (2015). Discovery of multiple hidden allosteric sites by combining Drug discovery using chemical systems biology: repositioning the safe medicine
Markov state models and experiments. Proc. Natl. Acad. Sci. U. S. A. Comtan to treat multi-drug and extensively drug resistant tuberculosis. PLoS Comput.
112, 2734–2739. doi: 10.1073/pnas.1417811112 Biol. 5:e1000423. doi: 10.1371/journal.pcbi.1000423 Kitchen, D. B., Decornez, H.,
Brown, A. S., and Patel, C. J. (2017). A standard database for drug Furr, J. R., and Bajorath, J. (2004). Docking and scoring in virtual screening for drug
repositioning. Sci. Data 4, 170029. doi: 10.1038/sdata.2017.29 discovery: methods and applications. Nat.
Chang, M., Smith, S., Thorpe, A., Barratt, M. J., and Karim, F. (2010). Evaluation of Rev. Drug Discov. 3, 935–949. doi: 10.1038/nrd1549
phenoxybenzamine in the CFA model of pain following gene expression studies Lamb, J., Crawford, E. D., Peck, D., Modell, J. W., Blat, I. C., Wrobel, M.
and connectivity mapping. Mol. Pain 6:56. doi: 10.1186/1744-8069-6-56 J., et al. (2006). The connectivity map: using gene-expression
Chen, Y.-C., Tolbert, R., Aronov, A. M., McGaughey, G., Walters, W. P., signatures to connect small molecules, genes, and disease. Science
and Meireles, L. (2016). Prediction of protein pairs sharing common 313, 1929–1935. doi: 10.1126/ science.1132939
active ligands using protein sequence, structure, and ligand similarity. Laurie, A. T. R., and Jackson, R. M. (2006). Methods for the prediction of
J. Chem. Inf. Model. 56, 1734–1745. doi: 10.1021/acs.jcim.6b00118 protein-ligand binding sites for structure-based drug design and virtual
Costa, F. F. (2014). Big data in biomedicine. Drug Discov. Today 19, 433– ligand screening. Curr. Protein Pept. Sci. 7, 395–406.
440. doi: 10.1016/j.drudis.2013.10.012 Li, Y. Y., An, J., and Jones, S. J. M. (2011). A computational approach to
Dakshanamurthy, S., Issa, N. T., Assefnia, S., Seshasayee, A., Peters, O. finding novel targets for existing drugs. PLoS Comput. Biol.
J., Madhavan, S., et al. (2012). Predicting new indications for approved 7:e1002139. doi: 10.1371/ journal.pcbi.1002139
drugs using a proteo-chemometric method. J. Med. Chem. 55, 6832– Liu, X., Ouyang, S., Yu, B., Liu, Y., Huang, K., Gong, J., et al. (2010).
6848. doi: 10.1021/ jm300576q PharmMapper server: a web server for potential drug target
Defranchi, E., De Franchi, E., Schalon, C., Messa, M., Onofri, F., identification using pharmacophore mapping approach. Nucleic Acids
Benfenati, F., et al. (2010). Binding of protein kinase inhibitors to Res. 38, W609–W614. doi: 10.1093/nar/gkq300
synapsin I inferred from pair-wise binding site similarity measurements. Manning, G., Whyte, D. B., Martinez, R., Hunter, T., and Sudarsanam, S.
PLoS ONE 5:e12214. doi: 10.1371/ journal.pone.0012214 (2002). The protein kinase complement of the human genome.
Ehrt, C., Brinkjost, T., and Koch, O. (2016). Impact of binding site Science 298, 1912–1934. doi: 10.1126/science.1075762
comparisons on medicinal chemistry and rational molecular design. J. Mervin, L. H., Afzal, A. M., Drakakis, G., Lewis, R., Engkvist, O., and Bender, A.
Med. Chem. 59, 4121–4151. doi: 10.1021/acs.jmedchem.6b00078 (2015). Target prediction utilising negative bioactivity data covering large
Gaulton, A., Hersey, A., Nowotka, M., Bento, A. P., Chambers, J., Mendez, chemical space. J. Cheminformatics 7, 51. doi: 10.1186/s13321-015-0098-y
D., et al. (2017). The ChEMBL database in 2017. Nucleic Acids Res. Mestres, J., Martín-Couce, L., Gregori-Puigjané, E., Cases, M., and Boyer, S.
45, D945–D954. doi: 10.1093/nar/gkw1074 (2006). Ligand-based approach to in silico pharmacology: nuclear receptor
Gregori-Puigjané, E., and Mestres, J. (2008). A ligand-based approach to profiling. J. Chem. Inf. Model. 46, 2725–2736. doi: 10.1021/ci600300k
mining the chemogenomic space of drugs. Comb. Chem. High Novac, N. (2013). Challenges and opportunities of drug repositioning.
Throughput Screen. 11, 669–676. Trends Pharmacol. Sci. 34, 267–272. doi: 10.1016/j.tips.2013.03.004
Hall, D. R., Kozakov, D., Whitty, A., and Vajda, S. (2015). Lessons from Sawada, R., Iwata, H., Mizutani, S., and Yamanishi, Y. (2015). Target-based
hot spot analysis for fragment-based drug discovery. Trends drug repositioning using large-scale chemical–protein interactome data. J.
Pharmacol. Sci. 36, 724–736. doi: 10.1016/j.tips.2015.08.003 Chem. Inf. Model. 55, 2717–2730. doi: 10.1021/acs.jcim.5b00330
Hu, G., and Agarwal, P. (2009). Human disease-drug network based on genomic Sgobba, M., Caporuscio, F., Anighoro, A., Portioli, C., and Rastelli, G.
expression profiles. PLoS ONE 4:e6536. doi: 10.1371/journal.pone.0006536 (2012). Application of a post-docking procedure based on MM-PBSA
Hu, Y., and Bajorath, J. (2013). Compound promiscuity: what can we learn and MM-GBSA on single and multiple protein conformations. Eur. J.
from current data? Drug Discov. Today 18, 644–650. doi: Med. Chem. 58, 431–440. doi: 10.1016/j.ejmech.2012.10.024
10.1016/j.drudis.2013.03. 002 Shameer, K., Glicksberg, B. S., Hodos, R., Johnson, K. W., Badgeley, M.
Huang, L., and Fu, L. (2015). Mechanisms of resistance to EGFR tyrosine kinase A., Readhead, B., et al. (2017). Systematic analyses of drugs and
inhibitors. Acta Pharm. Sin. B 5, 390–401. doi: 10.1016/j.apsb.2015.07.001 Inhester, disease indications in RepurposeDB reveal pharmacological, biological
T., Bietz, S., Hilbig, M., Schmidt, R., and Rarey, M. (2017). Index- and epidemiological factors influencing drug repositioning. Brief.
based searching of interaction patterns in large collections of protein-ligand Bioinform. doi: 10.1093/bib/bbw136 [Epub ahead of print].
interfaces. J. Chem. Inf. Model. 57, 148–158. doi: 10.1021/acs.jcim.6b00561 Stumpfe, D., and Bajorath, J. (2012). Exploring activity cli ffs in medicinal
Iskar, M., Zeller, G., Blattmann, P., Campillos, M., Kuhn, M., Kaminska, K. chemistry. J. Med. Chem. 55, 2932–2942. doi: 10.1021/jm201706b
H., et al. (2013). Characterization of drug-induced transcriptional Unterthiner, T., Mayr, A., Klambauer, G., Steijaert, M., Wegner, J. K.,
modules: towards drug repositioning and functional understanding. Ceulemans, H., et al. (2014). Deep Learning as an Opportunity in Virtual
Mol. Syst. Biol. 9, 662. doi: 10.1038/msb.2013.20 Screening. Available at: http://www.datascienceassn.org/sites/default/
Iwata, H., Sawada, R., Mizutani, S., and Yamanishi, Y. (2015). Systematic drug files/Deep%20Learning%20as%20an%20Opportunity%20in%20Virtual%
repositioning for a wide range of diseases with integrative analyses of 20Screening.pdf [accessed March 20, 2017].

Frontiers in Pharmacology | www.frontiersin.org 6 May 2017 | Volume 8 | Article 298

March-Vila et al. In silico Drug Repurposing

Vasudevan, S. R., Moore, J. B., Schymura, Y., and Churchill, G. C. (2012). Zhang, Y.-M., Cockerill, S., Guntrip, S. B., Rusnak, D., Smith, K.,
Shape-based reprofiling of FDA-approved drugs for the H1 histamine Vanderwall, D., et al. (2004). Synthesis and SAR of potent
receptor. J. Med. Chem. 55, 7054–7060. doi: 10.1021/jm 300671m EGFR/erbB2 dual inhibitors. Bioorg. Med. Chem. Lett. 14, 111–114.

Wang, Y., Bryant, S. H., Cheng, T., Wang, J., Gindulyte, A., Shoemaker, B. Conflict of Interest Statement: The authors declare that the research
A., et al. (2017). PubChem BioAssay: 2017 update. Nucleic Acids Res. was conducted in the absence of any commercial or financial relationships
45, D955–D963. doi: 10.1093/nar/gkw1118 that could be construed as a potential conflict of interest.
Wishart, D. S., Knox, C., Guo, A. C., Shrivastava, S., Hassanali, M.,
Stothard, P., et al. (2006). DrugBank: a comprehensive resource for in Copyright © 2017 March-Vila, Pinzi, Sturm, Tinivella, Engkvist, Chen and
silico drug discovery and exploration. Nucleic Acids Res. 34, D668– Rastelli. This is an open-access article distributed under the terms of the
D672. doi: 10.1093/nar/ gkj067 Creative Commons Attribution License (CC BY). The use, distribution or
Zhang, P., Wang, F., and Hu, J. (2014). Towards drug repositioning: a reproduction in other forums is permitted, provided the original author(s)
unified computational framework for integrating multiple aspects of or licensor are credited and that the original publication in this journal is
drug similarity and disease similarity. AMIA Annu. Symp. Proc. AMIA cited, in accordance with accepted academic practice. No use, distribution
Symp. 2014, 1258– 1267. or reproduction is permitted which does not comply with these terms.

Frontiers in Pharmacology | www.frontiersin.org 7 May 2017 | Volume 8 | Article 298