Send Orders for Reprints to reprints@benthamscience.ae 1381-6128/16 $58.00+.00 © 2016 Bentham Science Publishers Systems Pharmacology: A Unified Framework for Prediction of Drug-Target
Trang 1Send Orders for Reprints to reprints@benthamscience.ae
1381-6128/16 $58.00+.00 © 2016 Bentham Science Publishers
Systems Pharmacology: A Unified Framework for Prediction of Drug-Target Interactions
Duc-Hau Le*a and Ly Leb
a School of Computer Science and Engineering, Water Resources University, 175 Tay Son, Dong Da, Hanoi,
Vietnam; b School of Biotechnology, International University, Vietnam National University, Vietnam
Abstract: Background: Drug discovery is one important issue in medicine and pharmacology area Traditional
methods using target-based approach are usually time-consuming and ineffective Recently, the problems are
ap-proached in a system-level view and therefore it is called systems pharmacology This research field deals with the
problems in drug discovery by integrating various kinds of biomedical and pharmacological data and using
ad-vanced computational methods Ultimately, the problems are more effectively solved One of the most important
problem in systems pharmacology is prediction of drug-target interactions Methods: In this review, we are going
to summarize various computational methods for this problem Results: More importantly, we formed a unified
framework for the problem In addition, to study human health and disease in a more systematically and
effec-tively, we also presented an integrated scheme for a wider problem of prediction of disease-gene-drug associations
Conclusion: By presenting the unified framework and the integrated scheme, underlying computational methods for problems in systems
pharmacology can be understood and complex relationships among diseases, genes and drugs can be identified effectively
Keywords: Drug-target interaction, network-based approach, machine learning-based approach, drug-disease association, disease-gene
association, drug-gene-disease association
1 INTRODUCTION
The development of a drug from an original idea to the market
usually takes ten to seventeen years and costs about billion US
dol-lar on average [1, 2] In addition, several of them have been
with-drawn due to adverse/side effects The major reason is that the
drugs do not only interact to therapeutic targets but also to
off-targets In addition, drugs can interact with other drugs or chemicals
from food when they are used in the same patients Therefore, drug
discovery are needed to be considered at system-level, in which
many kinds of data for biomedical and pharmacological instances
are investigated simultaneously [3] In addition, computational
methods are also used to recognize interactions and mutual effects
among the instances Therefore, system pharmacology, which is a
combination between systems biology and pharmacology, becomes
promising approach to deal with this problem In which, systems
biology is the way to investigate interactions among cellular
com-ponents at system-level Therefore, systems pharmacology is a
biological and chemical approach for health and diseases [4-6]
More specifically, systems pharmacology studies how drugs work
(i.e., mode of action) at different levels such as whole human body,
organs, tissues, or cellular components [7] Computational methods
in systems pharmacology are usually used along with various kinds
of data such as –omics (genomics, transcriptomics, proteomics,
metabolomics, phenomics, interactomics) as well as
pharmacologi-cal data, etc.… [8] The final goals is to elucidate therapeutic
mechanisms of drugs (i.e., how drug work on different pathways,
cell types and tissues) and enable the process of drug discovery and
development to become more effective Based on these
computa-tional methods, many state-of-the-art technologies in modern
com-puter-aided drug design have been developed [9]
There is a number of problems in drug discovery such as
identi-fication of targets, drug repositioning, drug efficacy (e.g., safety,
*Address correspondence to this author at School of Computer Science and
Engineering, Water Resources University, 175 Tay Son, Dong Da, Hanoi,
Vietnam; Tel/Fax: +84-912324564; E-mails: hauldhut@gmail.com
adverse effects and responses) [10] Many computational methods have been proposed to solve these problems [11-16] In this study,
we focus on the most important problem in drug discovery, which
is prediction of novel drug-target interactions [17-19] To identify targets, experimental studies usually approach the problem based on either target or phenotype [20] Meanwhile, computational methods for this problem are very diverse, in which novel interactions are predicted based on known drug-target interactions using various kinds of supporting biomedical and pharmacological data [21, 22] Irrespectively to this diversity, computational methods are classi-fied into two main approaches: i) network-based, ii) machine learn-ing-based There has been a number of studies which review the computational methods for predictions of drug-target interactions [15, 23, 24] However, these studies have simply listed proposed methods but not yet proposed a unified framework of the computa-tional methods for the problem For instance, studies [23, 24] fo-cused on only machine learning-based methods Recently, Chen
et al [15] has reviewed both network-based as well as machine
learning-based methods However, this study only enumerated pro-posed methods but not yet propro-posed any general framework for this problem In this study, we are going to review the computational methods as well as propose a unified framework for the prediction
of drug-target interactions In addition, to study human health and disease in a more systematically and effectively, we present an integrated scheme for a wider problem of prediction of disease-gene-drug associations
2 A UNIFIED FRAMEWORK FOR PREDICTION OF DRUG-TARGET INTERACTIONS
Drug discovery by systems pharmacological approaches re-quires the integration of various kinds of data [8, 25] This is sys-tems biology and chemistry-based approach [5] for human health and diseases [4] Therefore, all computational methods integrate different kinds of data such as –omics as well as pharmacological data [26, 27] Many computational methods have been proposed for the drug discovery [8], however, a majority of them are based on
Duc-Hau Le
Trang 2networks of biomedical and pharmacological instances (e.g., drugs,
targets and diseases), and therefore called network pharmacology
[10, 28-32] The network-based methods start with construction of
heterogeneous networks of biomedical and pharmacological
in-stances by combining homogeneous ones (i.e., drug similarity
net-works, target similarity networks and phenotypic disease similarity
network) and bipartite ones (i.e., drug-target networks, drug-disease
networks and disease-gene networks) In addition, network-based
methods consider action modes of drugs in the context of
interac-tions among cellular components [28] After heterogeneous
net-works are constructed, computational methods are then proposed to
identify novel interactions between drugs and targets Besides the
network-based approaches, other major approaches are based on
machine learning techniques [23, 24] The machine learning-based
approaches also require integrating various kinds of data to build
feature vectors for biomedical and pharmacological instances as
well as calculate similarity matrices among the instances After that,
a suitable machine learning algorithm is proposed to build a predic-tive model for identification of novel drug-target interactions
In general, these two main approaches are both based on simi-larity measures because they are all under an assumption that chemically or pharmacologically similar drugs target similar target proteins [33] Network-based approaches use similarity measures to construct the heterogeneous networks and calculate relative similar-ity between candidate interactions and known ones Meanwhile, machine learning-approaches use them for building kernel matrices
as well as calculating similarity among feature vectors/interaction profiles In addition, three main types of data spaces are used in these approaches, they are genomic space (i.e., general proteins as well as target ones), chemical space (i.e., chemical structures of drugs) and pharmacological space (i.e., general phenotypes as well
as drug-affected ones) However, in order to build kernel matrices,
Fig (1) A unified framework for prediction of drug-target interactions
Trang 3Systems Pharmacology: A Unified Framework for Prediction Current Pharmaceutical Design, 2016, Vol 22, No 23 3571
similarity networks and feature vectors, many types of data have
been used such as chemical structure of drugs, sequence of proteins,
network of instances and text (e.g., ontologies and literature) This
requires different similarity measures to effectively estimate how
similar two instances are In addition, the proposed methods are not
only application of a network/machine learning-based algorithm on
a raw data, but they are a processes containing sequential steps such
as data preparation, model building, algorithm selection, model
assessment and result analysis Therefore, the categorization of the
studies into two main approaches (i.e., network-based and machine
learning-based) are simply based on algorithm selection An
exam-ple for the intersection between the two main approaches is that the
kernel matrices in machine learning-based methods can be
con-structed using kernel functions on graph/network In addition,
ma-chine learning-based methods can be started by building a bipartite
network such as known drug-target interaction networks [34] After
that, learning algorithms were used to predict novel ones [35-37]
Moreover, network-based methods are usually started with
con-struction of heterogeneous network of instances, then a
network-based algorithm is proposed to rank candidates (e.g., candidate
targets) based on their functional similarities to known ones (e.g.,
known drug-related targets) Therefore, this can be considered as a
learning process where the model is built totally based on labeled
data (e.g., known drug-related targets) (i.e., supervised learning) or
based on both labeled and unlabeled data (e.g., other targets in the
network) (i.e., semi-supervised learning) Another common issue in
the two main approaches is data representation In general, the
com-putational approaches represent data in two ways: i) drugs and
targets are considered as separate instances (e.g., given a known
drug and its known targets, predict novel targets of this drug), and
ii) drugs and targets are consider as pairs (e.g., given known
drug-target interactions, predict novel ones) Fig (1) shows a unified
framework for the two main approaches for the prediction of
drug-target interactions In the next section, we are going to summarize
some typical studies in each approach to clarify the point
2.1 Network-Based Approaches
Typical studies in this approach usually begin by constructing a
heterogeneous network of biomedical and pharmacological
in-stances, then a network-based algorithm is proposed to identify
novel drug-target interactions For instance, Chen et al [33] built a
heterogeneous network that combines a drug similarity network and
a target protein similarity network by known drug-target
interac-tions In which, the drug similarity network was constructed based
on similarity in chemical structures of drugs and the target protein
similarity network was constructed based on similarity among
se-quences of target proteins In addition, information of known
drug-target interactions was also embedded into these two similarity
networks After that, a random walk with restart (RWR) algorithm,
which was successfully used for prediction of disease-associated
genes [38] and microRNA [39] on a heterogeneous network, was
used to identify novel target proteins related to a given drug The
underlying assumption of this method (both for the construction of
the heterogeneous network as well as the selection of RWR
algo-rithm) is that chemically similar drugs target to similar target
pro-teins Therefore, this method was proven to outperform ones, which
was solely based on target protein similarity network Using the
same algorithm as in [33], however, Seal et al [40] additionally
used an extensive drug-target network and a drug similarity
net-work with links among drugs were built based on the molecular
similarity with chemical fingerprints Another network-based
method [41] was also proposed to predict novel drug-target
interac-tions They did not specifically build the heterogeneous network of
drugs and targets However, they still based chemical similarity
between drugs and sequence similarity between target proteins to
propose predictive drug-based similarity inference (DBSI) and
tar-get-based similarity inference (TBSI) models, respectively In
addi-tion, they proposed a network-based inference (NBI) model based
on topological similarity in the drug-target bipartite network Taken together, network-based methods are usually based on the similarity between drugs or between targets (i.e., in the form of similarity networks or similarity matrices) and a known drug-target interac-tion network
2.2 Machine Learning-Based Approaches
These approaches also mainly used the three data spaces as in network-based ones (i.e., genomic, chemical and pharmacological spaces) A main strategy of these methods is to build kernel matri-ces based on similarity among drugs and similarity among targets Basically, drugs are chemical compounds, which are in the form of graph structure, therefore graph-based kernel functions [42] are usually used to calculate kernel matrices Also, target proteins can
be represented as sequences, therefore string-based kernel functions can be used to calculate kernel matrices [43] There are two main approaches for machine learning-based methods for the prediction
of drug-target interactions, they are chemogenomics and pharmaco-genomics In which, the former is based on an assumption that chemically similar drugs usually target similar target proteins [35, 44-46] Meanwhile, the latter is based on another that phenotypi-cally similar drugs interact to similar target proteins [37, 47], in which, phenotypes of drugs can be represented by effects of drugs The built kernel matrices are then used for a learning model to pre-dict novel drug-target interactions
Most of the machine learning-based methods proposed for the problem are based on kernel-based supervised learning techniques
For instance, Jacob et al [45] used a product kernel (i.e., a
combi-nation of chemical structure-based kernel matrix for ligands and sequence-based kernel matrix for target proteins) In addition to
these kinds of data, Yamanishi et al [35] utilized topological
prop-erties of the drug-target network to build a kernel-based logistic regression prediction model By combining both chemical and
pharmacological data of drugs, Yamanishi et al [37] proposed to
use a supervised learning method on the bipartite drug-target
net-work In addition, Takarabe et al [48] also used the product Kernel
as in [45] and the kernel-based regression model as in [35], how-ever, they built pharmacological kernel matrix instead of the
chemical structure-based kernel matrix as in [35] Finally, Bleakley
et al [36] used the same kernel matrices as [48] in a bipartite local
models (BLM) on the bipartite drug-target network Based on the bipartite drug-target network, some studies built interaction profiles
of drugs and targets [49-51] For instance, studies [50, 51] proposed
a method, namely GIP, by building Gaussian kernels based on these interaction profiles After that, a regularized least square-based classifier was used to predict novel drug-target interactions [50] Besides, other machine learning models have been used for the problem such as conditional random field (CRF), a probabilistic graphical model [52] and Bayesian matrix factorization [53] Taken together, these machine learning-based methods mainly used simi-larity measures to build kernel matrices These kernel matrices were usually built separately for drugs and targets After that, they were combined for predictive models
Other supervised approaches are feature vector-based, in which, drug-target pairs were represented as feature vectors Based on features of drugs and targets extracted from drug and target-related data as well as known drug-target interactions, classifiers were then constructed to predict novel drug-target interactions There have been several learning models used for the problem such as k-nearest neighbor (kNN) [54], logistic regression [55, 56], support vector machines (SVM) [57-59] and ensemble methods [60, 61] Interest-ingly, most of these methods are kernel-based, where kernel matri-ces are calculated from feature vectors of training instanmatri-ces [56-59, 61]
A major limitation of the above supervised learning-based methods is that the predictive models were built based on only la-beled data (i.e., known drug-target interactions) However, this kind
Trang 4of data is very limited due to the cost of labeling process In
addi-tion, some of them are binary-classification models This means
that negative training instances (i.e., non-drug-target interactions)
have to be specified in training process However, there is no such
experimentally verified interactions in literature Therefore,
semi-supervised learning methods, which learn from both labeled and
unlabeled data (i.e., unknown drug-target interactions), have been
proposed to overcome this limitations [62-64] These methods also
utilized kernel/similarity matrices constructed from chemical and
genomic spaces as well as the bipartite drug-target network
3 AN INTEGRATED SCHEME FOR PREDICTION OF
DRUG-GENE-DISEASE ASSOCIATIONS
Approaches in systems pharmacology are based on various
kinds of data and computational methods In which, three main
entities (i.e., drug, disease and gene/target protein) should be
con-sidered in the same context [65-67] There have been many
individ-ual databases for each entity, but few for all of them such as
C2Maps [68], EU-ADR corpus [69] Therefore, there is a pressing
need to build such the databases In the future, these databases can
be formed and more comprehensive with results from experimental
studies as well as text mining techniques [3, 70, 71]
The relationships among these entities pose three challenges,
one of them is the prediction of drug-target interactions The two
remaining ones are prediction of disease-associated genes/proteins
and prediction of novel drug-disease associations (also known as
drug repositioning/repurposing) Although, the three problems have
their own goals in biomedicine and pharmacology, but they have
the same target in algorithmic view, which is prediction of novel
binary interactions/associations This is also a popular problem in
bioinformatics More importantly, they are all based on a
“guilt-by-association” assumption that functionally similar drugs/diseases
target/relate to functionally similar target proteins/genes,
respec-tively Among them, prediction of novel disease-gene associations
is very popular problem in biomedicine research and well-studied
by both network- and machine learning-based methods [72-80]
Therefore, we could use or adapt the methods used for this problem
for the prediction of drug-target interactions In addition, prediction
of novel drug-disease associations have been also studied for years [16, 32, 71, 81, 82]
As aforementioned, many computational methods have been proposed for prediction of drug-target interactions However, some
of them were built and assessed based on simple settings and ulti-mately they may not be used for practical problems Therefore, computational methods are needed to take intensive consideration
on the problem modeling as well as assessment methods [83] In addition, obviously, irrespective of the network- or machine learn-ing-based methods, the calculation of similarity among instances in different data spaces to construct similarity/kernel matrices and similarity networks is the most important step Data integra-tion/fusion from various spaces is also important since the problem
is considered in multi-dimension as well as systematically There-fore, we should also take consideration on selection of integration methods, in which kernel-based data fusion methods can be suitable ones since the data is mostly represented as kernel matrices Indeed,
Wang et al [84] is a pioneer applying kernel-based data fusion for
the prediction of drug-target interactions However, it should be noted that this data fusion method has been used popularly for pre-diction of disease-gene association for years [85-87] In addition, many algorithms have been proposed for kernel-based data fusion
[88] such as L p-norm MKL [89], which could be used for kernel-based problems Therefore, they could be used to improve the pre-diction of drug-target interactions
Besides, to meet practical problems, more suitable machine learning models should be considered for the prediction of drug-target interactions such as positive and unlabeled (PU) learning technique, which has been successfully applied in prediction of disease-gene associations [90] In addition, the combination be-tween the PU learning technique and kernel-based data fusion or ensemble methods could provide better way for the problem as did for prediction of disease-gene associations in [91] and [92], respec-tively For network-based approaches, the calculation of similarity between instances in constructed networks can be done with graph-based kernel functions [93, 94] Some of them have been used suc-cessfully for prediction of disease-gene associations Similarly,
Fig (2) An integrated scheme for prediction of drug-gene-disease associations
Trang 5Systems Pharmacology: A Unified Framework for Prediction Current Pharmaceutical Design, 2016, Vol 22, No 23 3573
these machine learning- and network-based methods could be used
to predict drug-target interactions more effectively
In parallel to proposal of computational methods and
construc-tion of databases, development of tools for predicconstruc-tion of novel
drug-target interactions is also needed [95, 96] There has been a
number of such the tools such as DINIES [97], iGPCR-Drug [98]
Also, many tools have been developed for prediction of
disease-gene associations [74, 99, 100] as well as several ones for
predic-tion of drug-disease associapredic-tions such as DRAR-CPI [101] and
PROMISCUOUS [102] Obviously, the above mentioned tools are
developed individually for each the problem of prediction
There-fore, an integrated tool predicting drug-gene-disease associations is
needed such as DT-Web [103], which can predict both drug-target
interactions and drug-disease associations
Taken together, to study human health and disease in a more
systematically and effectively, the drug-gene-disease associations
should be considered in the same context Therefore, all proposed
computational prediction methods, constructed databases and
de-veloped tools should meet this requirement Figure (2) shows an
integrated scheme for prediction of drug-gene-disease associations
CONCLUSION
After human genome project, it has been expected that several
new effective drugs will be developed Unfortunately, most of drug
targets are proteins which go through significant post translation
modification or involve in complex network Systems
pharmacol-ogy therefore becomes the modern approach for drug discovery and
development in post genomic era In this study, we have reviewed
various computational approaches proposed for this problem More
importantly, by analyzing the proposed methods, we recognized
that they are based on the same assumption that similar drugs target
similar target proteins Based on this assumption, they use
similar-ity measures to calculate similarsimilar-ity among instances in different
data spaces to construct similarity/kernel matrices and similarity
networks After that, these matrices and networks are used in
net-work- or machine learning-based computational models to predict
novel drug-target interactions To be more intuitively, we proposed
a unified framework of computational methods for the prediction of
drug-target interactions In addition, we extended the problem to
effectively study human health and diseases by considering the
relationships among drugs, gene/target proteins and diseases in the
same context Interestingly, we found that the drug-gene-disease
relationship pose three key challenges (i.e., prediction of
drug-target interactions, prediction of disease-associated genes/proteins
and prediction of novel drug-disease associations) with similar
underlying assumptions (i.e., guilt-by-association) and the same
algorithmic view (i.e., prediction of binary interactions) Therefore,
predictive models and data fusion methods could be used
ex-changeably among the three problems Similarly, we also presented
an integrated scheme for prediction of drug-gene-disease
associa-tions
Finally, with high-throughput technologies in next generation
sequencing, GWAS has yielded large amount of genomic data
Therefore, computational methods for the problems in drug
discov-ery should deal with this wealth data [104, 105] which open a wide
road for research in personalized medicine [106-108] In addition,
novel predicted drug-target interactions by computational methods
need to be experimentally verified in wet-laboratory
CONFLICT OF INTEREST
The authors confirm that this article content has no conflict of
interest
ACKNOWLEDGEMENTS
This research is funded by Vietnam National Foundation for
Science and Technology Development (NAFOSTED) under grant
number 102.01-2014.21
REFERENCES
[1] Ashburn TT, Thor KB Drug repositioning: identifying and devel-oping new uses for existing drugs Nat Rev Drug Discov 2004; 3: 673-83
[2] Hughes JP, Rees S, Kalindjian SB, Philpott KL Principles of early drug discovery Br J Pharmacol 2011; 162: 1239-49
[3] Yang Y, Adelstein SJ, Kassis AI Target discovery from data min-ing approaches Drug Discov Today 2009; 14: 147-154
[4] Somvanshi P, Venkatesh KV A conceptual review on systems biology in health and diseases: from biological networks to modern therapeutics Syst Synth Biol 2014; 8: 99-116
[5] Brown JB, Okuno Y Systems biology and systems chemistry: new directions for drug discovery Chem Biol 2012; 19: 23-8
[6] Robinson R From Complex Behavior to New Drugs: Chemoin-formatics Shows the Way PLoS Biol 2013; 11: e1001713 [7] Zhao S, Iyengar R Systems pharmacology: Network analysis to identify multiscale mechanisms of drug action Ann Rev Pharmacol Toxicol 2012; 52: 505-21
[8] Ou-Yang S-s, Lu J-y, Kong X-q, Liang Z-j, Luo C, Jiang H Com-putational drug discovery Acta Pharm Sin 2012; 33: 1131-40 [9] Dalkas GA, Vlachakis D, Tsagkrasoulis D, Kastania A, Kossida S State-of-the-art technology in modern computer-aided drug design Brief Bioinform 2013; 14: 745-52
[10] Chandra N, Padiadpu J Network approaches to drug discovery Expert Opin Drug Discov 2013; 8: 7-20
[11] Iorio F, Saez-Rodriguez J, Bernardo Dd Network based elucida-tion of drug response: from modulators to targets BMC Syst Biol 2013; 7: 139
[12] Ryall K, Tan A Systems biology approaches for advancing the discovery of effective drug combinations J Cheminform 2015; 7:
7
[13] Takarabe M, Shigemizu D, Kotera M, Goto S, Kanehisa M Net-work-based analysis and characterization of adverse drug-drug in-teractions J Chem Inf Model 2011; 51: 2977-85
[14] Cheng F, Zhao Z Machine learning-based prediction of drug–drug interactions by integrating drug phenotypic, therapeutic, chemical, and genomic properties J Am Med Informatics Assoc 2014; 21: e278-86
[15] Chen X, Yan CC, Zhang X, et al Drug–target interaction
predic-tion: databases, web servers and computational models Brief Bio-inform 2015; [Epub ahead of print]
[16] Li J, Zheng S, Chen B, Butte AJ, Swamidass SJ, Lu Z A survey of current trends in computational drug repositioning Brief Bioinform 2016; 17(1): 2-12
[17] Overington JP, Al-Lazikani B, Hopkins AL How many drug tar-gets are there? Nat Rev Drug Discov 2006; 5: 993-96
[18] Imming P, Sinning C, Meyer A Drugs, their targets and the nature and number of drug targets Nat Rev Drug Discov 2006; 5: 821-34 [19] Keiser MJ, Setola V, Irwin JJ, et al Predicting new molecular
targets for known drugs Nature 2009; 462: 175-81
[20] Schenone M, Dancik V, Wagner BK, Clemons PA Target identifi-cation and mechanism of action in chemical biology and drug dis-covery Nat Chem Biol 2013; 9: 232-40
[21] Chandra N Computational systems approach for drug target dis-covery Exp Opin Drug Discov 2009; 4: 1221-36
[22] Yamanishi Y Chemogenomic approaches to infer drug–target interaction networks In: Mamitsuka H, DeLisi C, Kanehisa M, Eds Data Mining for Systems Biology USA: Humana Press 2013;
pp 97-113
[23] Ding H, Takigawa I, Mamitsuka H, Zhu S Similarity-based ma-chine learning methods for predicting drug–target interactions: a brief review Brief Bioinform 2014; 15: 734-47
[24] Mousavian Z, Masoudi-Nejad A Drug–target interaction prediction via chemogenomic space: learning-based methods Exp Opin Drug Metabol Toxicol 2014; 10: 1273-87
[25] Iskar M, Zeller G, Zhao XM, van Noort V, Bork P Drug discovery
in the age of systems biology: the rise of computational approaches for data integration Curr Opin Biotechnol 2012; 23: 609-16 [26] Masoudi-Nejad A, Mousavian Z, Bozorgmehr J Drug-target and disease networks: polypharmacology in the post-genomic era In Silico Pharmacol 2013; 1: 17
[27] Leung EL, Cao Z-W, Jiang Z-H, Zhou H, Liu L Network-based drug discovery by integrating systems biology and computational technologies Brief Bioinform 2013; 14: 491-505
Trang 6[28] Berger SI, Iyengar R Network analyses in systems pharmacology
Bioinformatics 2009; 25: 2466-72
[29] Csermely P, Korcsmáros T, Kiss HJ, London G, Nussinov R
Structure and dynamics of molecular networks: a novel paradigm
of drug discovery: a comprehensive review Pharmacol Ther 2013;
138: 333-408
[30] Ainsworth C Networking for new drugs Nat Med 2011; 17:
1166-8
[31] Hopkins AL Network pharmacology: the next paradigm in drug
discovery Nat Chem Biol 2008; 4: 682-90
[32] Wu Z, Wang Y, Chen L Network-based drug repositioning Mol
BioSyst 2013; 9: 1268-81
[33] Chen X, Liu MX, Yan GY Drug-target interaction prediction by
random walk on the heterogeneous network Mol BioSyst 2012; 8:
1970-8
[34] Yildirim MA, Goh K-I, Cusick ME, Barabasi A-L, Vidal M
Drug-target network Nat Biotech 2007; 25: 1119-26
[35] Yamanishi Y, Araki M, Gutteridge A, Honda W, Kanehisa M
Prediction of drug–target interaction networks from the integration
of chemical and genomic spaces Bioinformatics 2008; 24: i232-40
[36] Bleakley K, Yamanishi Y Supervised prediction of drug–target
interactions using bipartite local models Bioinformatics 2009; 25:
2397-403
[37] Yamanishi Y, Kotera M, Kanehisa M, Goto S Drug-target
interac-tion predicinterac-tion from chemical, genomic and pharmacological data
in an integrated framework Bioinformatics 2010; 26: i246-54
[38] Li Y, Patra JC Genome-wide inferring gene-phenotype
relation-ship by walking on the heterogeneous network Bioinformatics
2010; 26: 1219-24
[39] Le D-H Disease phenotype similarity improves the prediction of
novel disease-associated microRNAs 2015 2nd National
Founda-tion for Science and Technology Development Conference on
In-formation and Computer Science (NICS) 2015; pp 76-81
[40] Seal A, Ahn Y-Y, Wild D Optimizing drug-target interaction
pre-diction based on random walk on heterogeneous networks J
Cheminform 2015; 7: 40
[41] Cheng F, Liu C, Jiang J, et al Prediction of drug-target interactions
and drug repositioning via network-based inference PLoS Comput
Biol 2012; 8: e1002503
[42] Mahé P, Ueda N, Akutsu T, Perret JL, Vert JP Graph kernels for
molecular structureactivity relationship analysis with support
vec-tor machines J Chem Inf Model 2005; 45: 939-51
[43] Saigo H, Vert J-P, Ueda N, Akutsu T Protein homology detection
using string alignment kernels Bioinformatics 2004; 20: 1682-9
[44] Faulon J-L, Misra M, Martin S, Sale K, Sapra R Genome scale
enzyme–metabolite and drug–target interaction predictions using
the signature molecular descriptor Bioinformatics 2008; 24:
225-33
[45] Jacob L, Vert J-P Protein-ligand interaction prediction: an
im-proved chemogenomics approach Bioinformatics 2008; 24:
2149-56
[46] Yabuuchi H, Niijima S, Takematsu H, et al Analysis of multiple
compound–protein interactions reveals novel bioactive molecules
Mol Syst Biol 2011; 7
[47] Campillos M, Kuhn M, Gavin A-C, Jensen LJ, Bork P Drug target
identification using side-effect similarity Science 2008; 321:
263-6
[48] Takarabe M, Kotera M, Nishimura Y, Goto S, Yamanishi Y Drug
target prediction using adverse event report systems: a
pharmaco-genomic approach Bioinformatics 2012; 28: i611-8
[49] Mei JP, Kwoh CK, Yang P, Li XL, Zheng J Drug–target
interac-tion predicinterac-tion by learning from local informainterac-tion and neighbors
Bioinformatics 2013; 29: 238-45
[50] van Laarhoven T, Nabuurs SB, Marchiori E Gaussian interaction
profile kernels for predicting drug–target interaction
Bioinformat-ics 2011; 27: 3036-43
[51] van Laarhoven T, Marchiori E Predicting drug-target interactions
for new drug compounds using a weighted nearest neighbor profile
PLoS One 2013; 8: e66952
[52] Yang F, Xu J, Zeng J Drug-target interaction prediction by
inte-grating chemical, genomic, functional and pharmacological data
Pac Symp Biocomput 2014; 148-59
[53] Gönen M Predicting drug–target interactions from chemical and
genomic kernels using Bayesian matrix factorization
Bioinformat-ics 2012; 28: 2304-10
[54] He Z, Zhang J, Shi XH, et al Predicting drug-target interaction
networks based on functional groups and biological features PLoS One 2010; 5: e9603
[55] Perlman L, Gottlieb A, Atias N, Ruppin E, Sharan R Combining drug and gene similarity measures for drug-target elucidation J Comput Biol 2011; 18: 133-45
[56] Tabei Y, Pauwels E, Stoven V, Takemoto K, Yamanishi Y Identi-fication of chemogenomic features from drug–target interaction networks using interpretable classifiers Bioinformatics 2012; 28: i487-94
[57] Tabei Y, Yamanishi Y Scalable prediction of compound-protein interactions using minwise hashing BMC Syst Biol 2013; 7: S3 [58] Cao D-S, Liu S, Xu QS, et al Large-scale prediction of drug–target
interactions using protein sequences and drug topological struc-tures Anal Chim Acta 2012; 752: 1-10
[59] Wang Y-C, Yang Z-X, Wang Y, Deng N-Y Computationally probing drug-protein interactions via support vector machine Lett Drug Des Discov 2010; 7: 370-8
[60] Yu H, Chen J, Xu X, et al A Systematic Prediction of multiple
drug-target interactions from chemical, genomic, and pharmacol-ogical data PLoS One 2012; 7: e37608
[61] Nanni L, Lumini A, Brahnam S A set of descriptors for identifying the protein–drug interaction in cellular networking Joretical Biol 2014; 359: 120-8
[62] Xia Z, Wu L-Y, Zhou X, Wong S Semi-supervised drug-protein interaction prediction from heterogeneous biological spaces BMC Systems Biol 2010; 4: S6
[63] Chen H, Zhang Z A Semi-supervised method for drug-target inter-action prediction with consistency in networks PLoS One 2013; 8: e62975
[64] Yu W, Yan Y, Liu Q, Wang J, Jiang Z Predicting drug–target interaction networks of human diseases based on multiple feature information Pharmacogenomics 2013; 14: 1701-7
[65] Lamb J, Crawford ED, Peck D, et al The connectivity map: Using
gene-expression signatures to connect small molecules, genes, and disease Science 2006; 313: 1929-35
[66] Sun PG The human drug–disease–gene network Inf Sci 2015; 306: 70-80
[67] Peng Gang S Co-controllability of drug-disease-gene network New J Phys 2015; 17: 085009
[68] Huang H, Wu X, Pandey R, et al C2Maps: a network
pharmacol-ogy database with comprehensive disease-gene-drug connectivity relationships BMC Genom 2012; 13: 1-14
[69] Van Mulligen EM, Fourrier-Reglat A, Gurwitz D, et al The
EU-ADR corpus: annotated drugs, diseases, targets, and their relation-ships J Biomed Inform 2012; 45: 879-84
[70] Frijters R, van Vugt M, Smeets R, van Schaik R, de Vlieg J, Alkema W Literature mining for the discovery of hidden connec-tions between drugs, genes and diseases PLoS Comput Biol 2010; 6: e1000943
[71] Andronis C, Sharma A, Virvilis V, Deftereos S, Persidis A Litera-ture mining, ontologies and information visualization for drug re-purposing Brief Bioinform 2011; 12: 357-68
[72] Piro RM, Di Cunto F Computational approaches to disease-gene prediction: rationale, classification and successes FEBS J 2012; 279: 678-96
[73] Barabasi A-L, Gulbahce N, Loscalzo J Network medicine: a net-work-based approach to human disease Nat Rev Genet 2011; 12: 56-68
[74] Oti M, Ballouz S, Wouters MA Web Tools for the Prioritization of Candidate Disease Genes Methods Mol Biol 2011; 760: 189-206 [75] Wang X, Gulbahce N, Yu H Network-based methods for human disease gene prediction Brief Funct Genom 2011; 10: 280-93 [76] Le D-H A novel method for identifying disease associated protein complexes based on functional similarity protein complex net-works Algorithms Mol Biol 2015; 10: 14
[77] Le D-H Network-based ranking methods for prediction of novel disease associated microRNAs Comput Biol Chem 2015; 58:
139-48
[78] Le DH, Nguyen MH Towards more realistic machine learning techniques for prediction of disease-associated genes USA: SOICT
2015 Proceeding of the Sixth International Symposium on Informa-tion and ComunicaInforma-tion Technology 2015; pp 116-20
[79] Le D-H, Xuan Hoai N, Kwon Y-K A Comparative Study of Clas-sification-Based Machine Learning Methods for Novel Disease Gene Prediction In: Nguyen V-H, Le A-C, Huynh V-N, Eds.,
Trang 7Systems Pharmacology: A Unified Framework for Prediction Current Pharmaceutical Design, 2016, Vol 22, No 23 3575
Knowledge and Systems Engineering USA: Springer International
Publishing 2015; pp 577-88
[80] Le D-H, Kwon Y-K Neighbor-favoring weight reinforcement to
improve random walk-based disease gene prioritization Comput
Biol Chem 2013; 44: 1-8
[81] Sanseau P, Koehler J Editorial: Computational methods for drug
repurposing Brief Bioinform 2011; 12: 301-2
[82] Dudley JT, Deshpande T, Butte AJ Exploiting drug–disease
rela-tionships for computational drug repositioning Brief Bioinform
2011; 12: 303-11
[83] Pahikkala T, Airola A, Pietilä S, et al Toward more realistic drug–
target interaction predictions Brief Bioinform 2015; 16: 325-37
[84] Wang Y-C, Zhang C-H, Deng N-Y, Wang Y Kernel-based data
fusion improves the drug–protein interaction prediction Comput
Biol Chem 2011; 35: 353-62
[85] De Bie T, Tranchevent L-C, Van Oeffelen LMM, Moreau Y
Ker-nel-based data fusion for gene prioritization Bioinformatics 2007;
23: i125-32
[86] Yu S, Tranchevent L-C, De Moor B, Moreau Y Gene prioritization
and clustering by multi-view text mining BMC Bioinform 2010;
11: 28
[87] Yu S, Falck T, Daemen A, et al L2-norm multiple kernel learning
and its application to biomedical data fusion BMC Bioinform
2010; 11: 309
[88] Gönen M, Alpaydin E Multiple kernel learning algorithms J
Mach Learn Res 2011; 999999: 2211-68
[89] Kloft M, Brefeld U, Sonnenburg S, Zien A Lp-norm multiple
kernel learning J Mach Learn Res 2011; 999999: 953-97
[90] Yang P, Li X-L, Mei J-P, Kwoh C-K, Ng S-K Positive-unlabeled
learning for disease gene identification Bioinformatics 2012; 28:
2640-7
[91] Mordelet F, Vert J-P ProDiGe: Prioritization Of Disease Genes
with multitask machine learning from positive and unlabeled
ex-amples BMC Bioinformatics 2011; 12: 389
[92] Yang P, Li X, Chua H-N, Kwoh C-K, Ng S-K Ensemble positive
unlabeled learning for disease gene identification PLoS One 2014;
9: e97079
[93] Ma X, Chen T, Sun F Integrative approaches for predicting protein
function and prioritizing genes for complex phenotypes using
pro-tein interaction networks Brief Bioinform 2013: bbt041
[94] Fouss F, Francoisse K, Yen L, Pirotte A, Saerens M An
experi-mental investigation of kernels on graphs for collaborative
recom-mendation and semisupervised classification Neural Netw 2012; 31: 53-72
[95] Farkas IJ, Korcsmáros T, Kovács IA, et al Network-based tools for
the identification of novel drug targets Sci Signal 2011; 4: pt3-pt3 [96] Huan T, Wu X, Chen JY Systems biology visualization tools for drug target discovery Exp Opin Drug Discov 2010; 5: 425-39 [97] Yamanishi Y, Kotera M, Moriya Y, Sawada R, Kanehisa M, Goto
S DINIES: drug–target interaction network inference engine based
on supervised analysis Nucleic Acids Res 2014; 42: W39-45 [98] Xiao X, Min J-L, Wang P, Chou K-C iGPCR-drug: a web server for predicting interaction between GPCRs and drugs in cellular networking PLoS One 2013; 8: e72234
[99] Le D-H, Kwon Y-K GPEC: A Cytoscape plug-in for random walk-based gene prioritization and biomedical evidence collection Comput Biol Chem 2012; 37: 17-23
[100] Tranchevent L-C, Capdevila FB, Nitsch D, et al A guide to web
tools to prioritize candidate genes Brief Bioinform 2010; 12:
22-32
[101] Luo H, Chen J, Shi L, et al DRAR-CPI: a server for identifying
drug repositioning potential and adverse drug reactions via the chemical–protein interactome Nucleic Acids Res 2011; 39:
W492-8
[102] von Eichborn J, Murgueitio MS, Dunkel M, Koerner S, Bourne PE, Preissner R PROMISCUOUS: a database for network-based drug-repositioning Nucleic Acids Res 2011; 39: D1060-6
[103] Alaimo S, Bonnici V, Cancemi D, Ferro A, Giugno R, Pulvirenti
A DT-Web: a web-based application for drug-target interaction and drug combination prediction through domain-tuned network-based inference BMC Syst Biol 2015; 9: S4
[104] Sanseau P, Agarwal P, Barnes MR, et al Use of genome-wide
association studies for drug repositioning Nat Biotech 2012; 30: 317-20
[105] Cao C, Moult J GWAS and drug targets BMC Genom 2014; 15: S5
[106] Li Y, Jones S Drug repositioning for personalized medicine Ge-nom Med 2012; 4: 1-14
[107] Reynolds KS Achieving the promise of personalized medicine Clin Pharmacol Ther 2012; 92: 401-5
[108] Fernald GH, Capriotti E, Daneshjou R, Karczewski KJ, Altman
RB Bioinformatics challenges for personalized medicine Bioin-formatics 2011; 27: 1741-8
The author has requested enhancement of the downloaded file All in-text references underlined in blue are linked to publications on ResearchGate The author has requested enhancement of the downloaded file All in-text references underlined in blue are linked to publications on ResearchGate