Networks and Algorithms in Heterogeneous Network-based Methods

for Drug-target Interaction Prediction: A Survey and Comparison

Shanglin Gao

, Zhixing Liu

2,* b

and Ying Hong Li

1,* c

Chongqing Key Laboratory of Big Data for Bio Intelligence, Chongqing University of Posts and Telecommunications,

Chongqing, China

The Second People's Hospital of Jiulongpo District, Chongqing, China

Corresponding author

Keywords: Drug-Target Interaction, Heterogeneous Network-Based Methods, Networks, Algorithms.

Abstract: A key step in drug discovery is the identification of drug-target interactions (DTIs). However, only a small

fraction of DTIs have been experimentally validated due to the time-consuming and expensive aspects of

experimental validation. To improve the efficiency of drug discovery, many computer-aided drug-target

prediction methods have been developed to guide experimental validation. There are numerous prediction

methods for DTIs, among which heterogeneous network-based methods do not depend on the 3D structures

of the targets or compound molecules and they avoid the shortcomings of machine learning methods for

negative training dataset selection, exhibiting greater advantages than other methods. Currently, although

many reviews of drug-target prediction methods exist, only a few of them have addressed network-based

methods, and they have not been compared in terms of the heterogeneous networks and algorithms used.

Therefore, this paper presents a review of the heterogeneous network-based methods for DTI prediction,

compares the differences in the prediction performance of different heterogeneous networks and algorithms

from the perspective of the networks and algorithms used by these methods, and provides suggestions for

the selection of heterogeneous networks and algorithms.

1 INTRODUCTION

Drug-target interactions (DTIs) can be

experimentally validated by wet-laboratory methods

(e.g., affinity chromatography, etc.) (Bi et al. 2015).

However, these experiments are time-consuming

and costly, and large-scale validation is not possible.

Therefore, predicting DTIs by computer-assisted

methods will significantly reduce the scope of

experimental validation and improve the efficiency

of drug discovery. With the rapid increase in the

number of compounds (Kim et al. 2021), the

proportion of compound molecules with known

target characteristics and drug effects has decreased.

In addition, researchers have accumulated a large

amount of information on compounds, proteins, and

interactions to construct larger datasets, making it

https://orcid.org/0000-0003-4735-7245

https://orcid.org/0000-0002-1351-9149

https://orcid.org/0000-0003-3629-519X

possible to develop more accurate and efficient

methods to predict DTIs.

DTI prediction has multiple applications, such as

facilitating drug discovery (Chen Z. H. et al. 2020),

drug repositioning (Chen Z. H. et al. 2020), and drug

side-effect prediction (Pliakos & Vens 2020). The

drug discovery process is long, has a low success

rate, and consumes significant resources. It is

estimated that it takes approximately 10–15 years to

develop a new drug, consuming an average of $1.8

billion (Paul et al. 2010). Currently, the main reason

why the vast majority of the compounds that have

been discovered are not used as drugs is that the

interaction of these compounds with proteins is

unknown. Therefore, a computer-aided approach to

predict compound-protein interactions would have

the potential to significantly narrow the drug search

space and improve the efficiency of drug discovery.

Drug repositioning is a research strategy for new

uses outside the scope of the original medical

indication for a marketed drug or a clinical trial drug

(Ashburn & Thor 2004). The safety of approved

Gao, S., Liu, Z. and Li, Y.

Networks and Algorithms in Heterogeneous Network-based Methods for Drug-target Interaction Prediction: A Survey and Comparison.

DOI: 10.5220/0011230500003438

In Proceedings of the 1st International Conference on Health Big Data and Intelligent Healthcare (ICHIH 2022), pages 67-75

ISBN: 978-989-758-596-8

drugs or clinical trial drugs has been widely

confirmed due to the extensive clinical trials they

have undergone. Since the outbreak of the COVID-

19 pandemic, drug repositioning has become a

method for the rapid development of potent anti-

COVID-19 drugs. Drug repositioning studies can

either directly predict drug molecules for treating a

disease or to screen potential drug molecules by DTI

prediction in the context of identifying therapeutic

targets. DTI prediction has become an important

research direction in drug repositioning. The

combination of a drug with a therapeutic target may

produce therapeutic effects, while the combination

of a drug with other targets may produce side

effects. Drug side effects have become a major cause

of drug clinical trial failure (Pliakos & Vens 2020).

Therefore, predicting possible drug side effects by

DTI at the preclinical study stage will help in

selecting more suitable drug molecules for clinical

trials.

Therefore, DTI prediction will be very useful in

drug development. Prediction methods for DTIs are

generally divided into three categories (Sachdev &

Gupta 2019): ligand-based methods, docking

methods, and chemical genomics methods. Ligand-

based methods were developed based on the idea

that similar molecules usually bind to similar protein

targets and display similar properties (Jacob & Vert

2008). Docking methods use simulations of the

three-dimensional structures of proteins and drugs to

predict whether they will interact with each other

(Nagamine et al. 2009). Chemogenomic approaches

use information from both drugs and proteins for

interaction predictions (Zhao et al. 2019).

Heterogeneous network-based methods are the

best type of chemical genomics methods for

prediction (Ezzat et al. 2019), which do not depend

on the 3D structure of targets and compound

molecules or avoid the defects of negative data

selection of machine learning methods, showing

greater advantages than other methods. The methods

based on heterogeneous networks can be generally

classified into network inference (Saint-Antoine &

Singh 2020; Cheng et al. 2012), network

propagation (Engin et al. 2014), and matrix

decomposition (Hodos et al. 2016; Abbou et al.

2021) (Fig. 1). Several review articles have been

published about the prediction methods for DTIs,

which also contain a summary of the network-based

prediction methods (Wu et al. 2018). However, these

reviews do not provide a systematic comparison of

the heterogeneous networks and algorithms used by

these prediction algorithms. Therefore, this thesis

reviews recent heterogeneous network-based

forecasting methods for DTIs and proposes

recommendations for heterogeneous network

construction and algorithm selection after a

systematic comparison of the heterogeneous

networks and algorithms used.

Figure 1: The application and classification of DTIs.

ICHIH 2022 - International Conference on Health Big Data and Intelligent Healthcare

2 DTIs PREDICTIONS METHODS

This review summarizes the newly published

prediction methods for DTIs in recent years and

classifies them into the following categories:

network propagation, network inference, and matrix

factorization (Table 1).

Network propagation is a common approach

used to analyze heterogeneous networks, and a

variety of DTI prediction tools have been developed

based on this approach. NRWRH (Chen et al. 2012)

is a large-scale method for predicting DTIs

constructed by Chen et al. using a restarting random

walk algorithm under the assumption that similar

drugs usually have similar targets. This method

integrates three different networks (a protein

similarity network, a drug similarity network, and a

drug-target interaction network) into a “drug-target”

heterogeneous network. NRWRH was compared

with traditional supervised or semisupervised

methods such as NRWR (Chen et al. 2012), RWRH

(Li & Patra 2010), and RWR (Camoglu et al. 2005;

Kohler et al. 2008), which makes full use of

network-based information to achieve random

walking on the “drug-target” heterogeneous network

and improve the accuracy of predicting DTIs, but the

method still has certain shortcomings, such as the

problem of randomness, which is mainly caused by

the choice of the initial probability (Ganegoda et al.

2015). LPMIHN (Yan et al. 2016) is a label

propagation method optimized by Yan et al. based

on the NRWRH method. Its “drug-target”

heterogeneous network consists of a drug similarity

network, a target similarity network, and a drug-

target interaction network. Compared with NRWRH,

LPMIHN used a label propagation algorithm on the

constructed “drug-target” heterogeneous network to

infer potential DTIs, which reduces the network

sparsity problem caused by rare drug-target

interactions and further improves the prediction

accuracy. DTINet (Luo et al. 2017) is a

computational prediction pipeline developed by Luo

et al. that integrates multiple drug-related

information. Particularly, this method integrated six

different networks (including a drug-protein

interaction network, a protein similarity network, a

protein-disease association network, a drug-disease

association network, a drug similarity network, and a

drug-side effect association network) into the “drug-

disease-target-side effect” heterogeneous networks

and utilized the restart random walk algorithm to

accurately explain the topological characteristics of

each node in this heterogeneous network. In

addition, in experiments, Luo et al. verified the

novel interaction relationship between the three

drugs and the cyclooxygenase protein predicted by

DTINet and proved the new potential application of

these cyclooxygenase inhibitors in the prevention of

inflammatory diseases. Compared with HNM (Wang

et al. 2014), BLMNII (Mei et al. 2013), NetLapRLS

(Xia Z. et al. 2010), CMF (Xia L. Y. et al. 2019),

DTINet had a better predictive effect, which was

6.9% and 5.9% higher. Shahreza et al. developed a

semisupervised machine learning approach, Heter-

LP, using a label propagation algorithm on the

“drug-target-disease” heterogeneous network (Lotfi

Shahreza et al. 2019). The network of this approach

consists of a drug-disease association network, a

drug-target interaction network, and a disease-target

association network. In particular, Shahreza et al.

applied Heter-LP to analyze innovative putative

drug-disease, drug-target, and disease-target

relationships, including cosyntropin (drug) and

DHCR7, IGF1R, MC1R, MAP3K3, and TOP2A

(protein targets), for a rare disease adrenocortical

carcinoma (ACC). Heter-LP provided a new way for

the treatment of ACC (Lotfi Shahreza et al. 2017).

DHLP-1 (Maleki et al. 2020) and DHLP-2 (Maleki

et al. 2020), with two distributed label propagation

methods based on the “drug-target-disease”

heterogeneous network developed by Maleki et al.

Its heterogeneous network consists of a drug-disease

association network, a drug-target interaction

network, and a disease-target association network.

Compared with the two nondistributed methods,

MINProp (Lotfi Shahreza et al. 2017) and Heter-LP,

the two methods had superior results in terms of

running time and accuracy.

Network-based inference (NBI) is another

common approach to analyze heterogeneous

networks and it is frequently used in the prediction

methods for DTIs. HGBI (Wang et al. 2013) is a

new heterogeneous network-based inference method

proposed by Wang et al. This method constructs a

“drug-target” heterogeneous network by the known

drug-target interaction network, a drug similarity

network, and a target similarity network, and it

predicts DTIs based on this heterogeneous network.

Its prediction accuracy was improved compared with

NBI (Cheng et al. 2012) and BLM (Bleakley &

Yamanishi 2009). Wang et al. developed TL_HGBI

(Wang et al. 2014), which adopts the guilt-by-

association principle to integrate five networks

(including a disease similarity network, a drug-

disease association network, a drug similarity

network, a drug-target interaction network, and a

target similarity network) into the “drug-target-

disease” heterogeneous network. It optimized the

Networks and Algorithms in Heterogeneous Network-based Methods for Drug-target Interaction Prediction: A Survey and Comparison

HGBI method, particularly compared with other

methods. When the heterogeneous network model

was changed or the iterative algorithm was updated,

TL_HGBI could not only automatically construct a

new drug-target relationship network but also

automatically add drug-target information for drug-

disease relationship prediction. DT hybrid (domain

tuned-hybrid) (Alaimo et al. 2013) is an NBI

recommendation method based on heterogeneous

networks developed by Alaimo et al., integrating

NBI and Hybrid (Alaimo et al. 2013) tools. The

“drug-target” heterogeneous network of this method

includes a drug similarity network, a target

similarity network, and a drug-target interaction

network. Different from the traditional NBI

recommendation method, the DT hybrid takes into

account the important characteristics of the drug

target domain (Alaimo et al. 2015). SDTNBI (Wu et

al. 2017) is an NBI method based on a “drug-

substructure-target” heterogeneous network

developed by Wu et al. The heterogeneous network

consists of a new chemical entity-substructure

network, a substructure-drug network, and a drug-

target interaction network. This method prioritizes

potential targets of old drugs, failed drugs, and new

chemical entities and combines network and

chemical information to establish relationships

between new chemical entities and known DTI

networks. The advantage of SDTNBI is that it can

predict potential targets of new chemical entities,

whereas traditional network-based methods cannot.

The matrix factorization method can solve the

data sparsity problem well with better prediction

accuracy and has been widely used in the prediction

of DTIs. Liu et al. built a “drug-target”

heterogeneous network by integrating a drug

similarity network, a target similarity network, and a

drug-target interaction network while using the

matrix factorization method to develop NRLMF

(Liu et al. 2016). This method used the

neighborhood regularization logistic matrix

factorization algorithm to establish the interaction

probability model between the drug and the target, in

which the attributes of the drug and the target were

represented by the drug-specific and target-specific

potential vectors, respectively. The average AUC

and AUPR values of NRLMF in the gold standard

dataset are better than those of NetLapRLS (Xia Z.

et al. 2010), BLM-NII (Mei et al. 2013), WNN-GIP

(van Laarhoven & Marchiori 2013), KBMF2K

(Gonen 2012), CMF (Xia L. Y. et al. 2019). KMDR

(Kuang Q. F. et al. 2017) is a heterogeneous network

method based on the kernel matrix reduction

dimension algorithm developed by Kuang et al. The

“drug-target” heterogeneous

Table 1: Drug-target interaction predictions methods.

Name Networks

Algorithm

classification

Algorithms

Datasets for network

construction

Ref

NRWRH

drug-target (protein-protein +

drug-drug + drug-target)

Network

propagation

Random walk with

restarts (RWR)

DrugBank, KEGG,

SuperTarget,

Yamanishi et al.

(Yamanishi et al. 2008)

(Chen et

al. 2012)

HGBI

drug-target (drug-drug + target-

target + drug-target)

Network

inference

Network inference

Sophic

Integrated Druggable

Genome Database

(Sophic 2012), OMIM,

DrugBank, InterPro

(Hunter et al. 2009)

(Wang et

al. 2013)

TL_HGBI

drug-target-disease (disease-

disease + disease-drug + drug-

drug + drug-target + target-

target)

Network

inference

Triple layer

heterogeneous

graph based

inference

DrugBank, Sophic

Integrated Druggable

Genome Database

(Sophic 2012), OMIM,

Gottlieb et al. (Gottlieb

et al. 2011)

(Wang et

al. 2014)

DT-Hybrid

drug-target (drug-drug + target-

target + drug-target)

Network

inference

Bipartite network

projection

DrugBank,

Yamanishi et al.

(Yamanishi et al. 2008)

(Alaimo

et al.

2013)

DASPfind

drug-target (drug-drug + target-

target + drug-target)

Network path

analysis

Simple paths

finding

DrugBank, KEGG,

SuperTarget, BRENDA,

Yamanishi et al.

(Yamanishi et al. 2008)

(Ba-Alawi

et al.

2016)

NRLMF

drug-target (drug-drug + target-

target + drug-target)

Matrix

factorization

Neighborhood

regularized logistic

matrix factorization

Matador, ChEMBL,

DrugBank, KEGG,

SuperTarget, BRENDA

(Liu et al.

2016)

LPMIHN

drug-target (drug-drug + target-

target + drug-target)

Network

propagation

Label propagation

ChEMBL, DrugBank,

KEGG, SuperTarget,

(Yan et al.

2016)

ICHIH 2022 - International Conference on Health Big Data and Intelligent Healthcare

BRENDA

KMDR

drug-target (drug-drug + target-

target + drug-target)

Matrix

factorization

Kernel matrix

dimension

reduction

DrugBank, KEGG,

UniProt

(Kuang Q.

F. et al.

2017)

DTINet

drug-disease-target-side effect

(drug-protein + protein-protein

+ protein-disease + disease-

drug + drug-drug + drug-side

effect)

Network

propagation

RWR and diffusion

component analysis

DrugBank, HPRD, CTD,

SIDER

(Luo et al.

2017)

DNILMF

drug-target (drug-drug + target-

target + drug-target)

Matrix

factorization

A dual-network

integrated logistic

matrix factorization

DrugBank, KEGG,

BRENDA, SuperTarget,

COMPOUND

(Hao et al.

2017)

GRMF

drug-target (drug-drug + target-

target + drug-target)

Matrix

factorization

Graph regularized

matrix factorization

Yamanishi et al.

(Yamanishi et al. 2008)

(Ezzat et

al. 2017)

SDTNBI

substructure-drug-target

(drug-substructure + drug-

target +

new chemical entity-

substructure)

Network

inference

Network inference

ChEMBL, DrugBank,

BindingDB

(Wu et al.

2017)

Heter-LP

drug-target-disease (drug-

disease +

drug-target + disease-target)

Network

propagation

Label propagation DrugBank, SuperTarget

(Lotfi

Shahreza

et al.

2019)

DHLP-1

DHLP-2

drug-target-disease (drug-

disease +

drug-target + disease-target)

Network

propagation

Label propagation

Yamanishi et al.

(Yamanishi et al. 2008)

(Maleki et

al. 2020)

iDrug

drug-target-disease (drug-target

+ drug-disease + drug-drug +

targe

-target + disease-disease)

Matrix

factorization

Matrix factorization

CTD, Gottlieb et al.

(Gottlieb et al. 2011)

(Chen H.

et al.

2020)

network of KMDR consists of a drug similarity

network, a target similarity network, and a drug-

target interaction network. KMDR can reduce the

prediction bias, and it has a better DTI performance

than the regularized least squares classifier (RLS)

(Kuang Q. et al. 2014) and a semisupervised link

prediction classifier (SLP) (Kuang Q. et al. 2014).

DNILMF (Hao et al. 2017) is a dual-network

integrated logistic matrix factorization algorithm

developed by Hao et al., and its “drug-target”

heterogeneous network consists of a drug similarity

network, a target similarity network, and a drug-

target interaction network. This method used a

domain regularization logistic matrix factorization

algorithm, which was optimized based on NRLMF,

to improve the drug-target prediction accuracy, and

its prediction results had higher AUC and AUPR

values than NRLMF. Ezzat et al. developed a

network-based regularized matrix decomposition

tool, GRMF (Ezzat et al. 2017), whose “drug-target”

heterogeneity network consists of a drug similarity

network, a target similarity network, and a drug-

target interaction network.In addition, this method

took into account the situation in which many non-

occurring edges in the network were unknown or

missing cases and it added edges with intermediate

interaction probability scores in the preprocessing

step to improve the prediction results of the new

drugs and new targets. As a result, GRMF

performed very well in predicting the left-out

interactions. Chen et al. integrated a drug-target

interaction network, a drug-disease association

network, a drug similarity network, a disease

similarity network, and a target similarity network to

form the “drug-disease-target” heterogeneous

network and thus developed the iDrug (Chen H. et

al. 2020) method. This method utilized a matrix

factorization method to connect a drug-disease

association network and a drug-target interaction

network through drugs. MBiRW has better drug-

target prediction and drug-disease prediction

performance than TH_HGBI, and it can also identify

new drug-miRNA interactions.

In addition to the above three types of DTI

prediction methods, there are other methods, such as

network path analysis. DASPfind (Ba-Alawi et al.

2016) is a network path analysis method based on a

heterogeneous network developed by Ba-Alawi et al.

Its “drug-target” heterogeneous network consists of

a drug similarity network, a target similarity

network, and a drug-target interaction network.

Compared with the other methods, the advantage of

this method is that it can better predict DTIs with

unknown targets or drugs with fewer targets and it

has a better prediction performance than HGBI, DT-

Hybrid, and NRWRH.

Networks and Algorithms in Heterogeneous Network-based Methods for Drug-target Interaction Prediction: A Survey and Comparison

3 RESULTS AND DISCUSSION

3.1 Comparison of DTIs Prediction

Methods

Common algorithm evaluation methods include

independent dataset testing, ab initio prediction,

leave-one-out verification, and external dataset

verification, and the most commonly used cross-

validation is the tenfold cross-validation method,

which is widely used in the evaluation of algorithm

accuracy. The AUC value is the area under the ROC

curve, which can usually indicate the overall

performance of the algorithm, and it can be used to

compare the relative performance of different

algorithms, with larger values indicating better

algorithm performance (Sing et al. 2005). The PR

curve (precision recall curve) shows the relationship

between precision and recall. In most of the

literature, the indicators used to evaluate the

prediction performance of the algorithm are the area

under the curve (AUC) and the area under the

precision recall curve (AUPR) (Nascimento et al.

2016).

We collected the AUC and AUPR values of

more than a dozen methods, including NRWRH,

DT-Hybrid, DHLP, etc. on the gold standard dataset

(Lotfi Shahreza et al. 2018) (http://web.kuicr.kyoto-

u.ac.jp/supp/yoshi/drugtarget/), which was divided

into four parts (Yamanishi et al. 2008): enzyme, ion

channel, GPCR, and nuclear receptor (Table 2).

As shown in Table 2, DT-Hybrid, LPMIHN,

DNILMF, MINProp, NRLMF, and SDTNBI have

high AUC values on the same benchmark datasets,

and their AUC values on the four parts of

benchmark datasets are above 90%. In particular,

DT-Hybrid has a high prediction accuracy of DTIs

with AUC values of approximately 99% on the four

types of benchmark datasets. In addition, LPMIHN

has the highest AUPR value and has better

application prospects.

3.2 Comparison of Heterogeneous

Networks and Algorithms

In addition to the above comparison of the

performance of the DTI prediction methods through

the AUC and AUPR values, this paper also provided

statistics and comparisons of the effects of different

heterogeneous networks and different algorithms on

the prediction results (Table 3).

The comparisons shown in Table 3 indicate that

most of the methods with higher accuracy in

predicting DTIs used “drug-target” heterogeneous

networks constructed by a drug similarity network, a

target similarity network, and a drug-target

interaction network. The “drug-disease-target”

heterogeneous network constructed by adding

disease information did not contribute significantly

to an improvement of the prediction accuracy. Using

the same “drug-target” heterogeneous network, the

prediction accuracy of DNILMF and NRLMF using

the logistic matrix-based decomposition method was

higher than that of GRMF using only the matrix

decomposition method, and the AUPR value

increased from 76.3% to more than 98%. Therefore,

logistic matrix factorization was chosen as superior

for the prediction method of DTIs based on

heterogeneous networks. In addition, among the

methods using “drug-target” heterogeneous

networks, network propagation methods and

network inference methods were used for better

prediction.

Table 2: Reported AUC and AUPR on gold standard datasets in literature.

Method

Enzyme

Ion channel GPCR Nuclear receptor

Ref

AUC AUPR AUC AUPR AUC AUPR AUC AUPR

NRWRH 0.953 - 0.971 - 0.945 - 0.867 - (Chen et al. 2012)

DT-

Hybrid

0.999

- 0.997 -

0.999

1.000

(Eslami

Manoochehri &

Nourani 2020)

SDTNBI 0.958 - 0.971 - 0.966 - 0.932 - (Wu et al. 2017)

DASPfin

0.929 - 0.907 - 0.881 - 0.853 -

(Eslami

Manoochehri &

Nourani 2020)

HGBI 0.916 - 0.889 - 0.913 - 0.876 -

(Eslami

Manoochehri &

Nourani 2020)

ICHIH 2022 - International Conference on Health Big Data and Intelligent Healthcare

NRLMF 0.987 0.892 0.989 0.906 0.969 0.749 0.950 0.728 (Liu et al. 2016)

LPMIHN

0.999

0.929

0.998

0.961

0.998

0.973

0.996

0.970

(Yan et al. 2016)

GRMF - 0.763 - 0.745 - 0.567 - 0.423 (Ezzat et al. 2017)

DHLP-1 - - - - 0.976 0.766 - - (Maleki et al. 2020)

DLHP-2 - - - - 0.955 0.956 - - (Maleki et al. 2020)

HeterLP - - - - 0.967 0.796 - - (Maleki et al. 2020)

DNILMF 0.989 0.922 0.990 0.938 0.975 0.821 0.955 0.751 (Hao et al. 2017)

4 CONCLUSIONS

In this paper, we systematically reviewed the

heterogeneous network-based prediction methods

for DTIs, and the statistical analysis of the

heterogeneous networks showed that most of the

DTI prediction methods used the “drug-target”

heterogeneous network, which was comprised of a

drug similarity network, a target similarity network,

and a drug-target interaction network. In terms of the

algorithm selection methods, network inference,

network propagation and matrix factorization were

used for the prediction of DTIs. By comparing the

performance of these DTI methods against the gold

standard dataset, DT-Hybrid and LPMIHN were

found to have the best prediction performance. By

comparing the heterogeneous networks and

algorithms against the gold standard dataset, it was

found that the method using “drug-target”

heterogeneous networks had better prediction

performance and that the triple-layer heterogeneous

networks constructed by adding disease information

were of limited use in improving the prediction

accuracy. Among the “drug-target” heterogeneous

networks, network propagation and network

inference methods were found to have better

prediction performance.

ACKNOWLEDGEMENTS

This work was funded by the Natural Science

Foundation of Chongqing (cstc2019jcyj-

msxmX0271); the Science and Technology Research

Program of Chongqing Municipal Education

Commission (KJQN202100642).

REFERENCES

Abbou, S. I., Bouziane, H., & Chouarfia, A. (2021).

Logistic matrix factorisation and generative

adversarial neural network-based method for

predicting drug-target interactions. Mol Divers, 25 (3),

1497-1516.

Alaimo, S., Bonnici, V., Cancemi, D., et al. (2015). DT-

Web: a web-based application for drug-target

interaction and drug combination prediction through

domain-tuned network-based inference. BMC Syst

Biol, 9 Suppl 3, S4.

Alaimo, S., Pulvirenti, A., Giugno, R., et al. (2013). Drug-

target interaction prediction through domain-tuned

network-based inference. Bioinformatics, 29 (16),

2004-2008.

Ashburn, T. T., & Thor, K. B. (2004). Drug repositioning:

identifying and developing new uses for existing

drugs. Nature Reviews Drug Discovery, 3 (8), 673-

683.

Ba-Alawi, W., Soufan, O., Essack, M., et al. (2016).

DASPfind: new efficient method to predict drug-target

interactions. Journal of Cheminformatics, 8, 15.

Bi, C., Beeram, S., Li, Z., et al. (2015). Kinetic analysis of

drug-protein interactions by affinity chromatography.

Drug Discov Today Technol, 17, 16-21.

Bleakley, K., & Yamanishi, Y. (2009). Supervised

prediction of drug-target interactions using bipartite

local models. Bioinformatics, 25 (18), 2397-2403.

Camoglu, O., Can, T., Singh, A. K., et al. (2005). Decision

tree based information integration for automated

protein classification. Journal of Bioinformatics and

Computational Biology, 3 (3), 717-742.

Chen, H., Cheng, F., & Li, J. (2020). iDrug: Integration of

drug repositioning and drug-target prediction via

cross-network embedding. PLoS Computational

Biology, 16 (7), e1008040.

Chen, X., Liu, M. X., & Yan, G. Y. (2012). Drug-target

interaction prediction by random walk on the

heterogeneous network. Molecular Biosystems, 8 (7),

1970-1978.

Chen, Z. H., You, Z. H., Guo, Z. H., et al. (2020).

Prediction of Drug-Target Interactions From Multi-

Molecular Network Based on Deep Walk Embedding

Model. Front Bioeng Biotechnol, 8, 338.

Networks and Algorithms in Heterogeneous Network-based Methods for Drug-target Interaction Prediction: A Survey and Comparison

Cheng, F., Liu, C., Jiang, J., et al. (2012). Prediction of

drug-target interactions and drug repositioning via

network-based inference. PLoS Computational

Biology, 8 (5), e1002503.

Engin, H. B., Gursoy, A., Nussinov, R., et al. (2014).

Network-based strategies can help mono- and poly-

pharmacology drug discovery: a systems biology

view. Current Pharmaceutical Design, 20 (8), 1201-

1207.

Eslami Manoochehri, H., & Nourani, M. (2020). Drug-

target interaction prediction using semi-bipartite graph

model and deep learning. BMC Bioinformatics, 21

(Suppl 4), 248.

Ezzat, A., Wu, M., Li, X. L., et al. (2019). Computational

prediction of drug-target interactions using

chemogenomic approaches: an empirical survey. Brief

Bioinform, 20 (4), 1337-1357.

Ezzat, A., Zhao, P., Wu, M., et al. (2017). Drug-Target

Interaction Prediction with Graph Regularized Matrix

Factorization. IEEE/ACM Transactions on

Computational Biology Bioinformatics, 14 (3), 646-

656.

Ganegoda, G. U., Li, M., Wang, W., et al. (2015).

Heterogeneous network model to infer human disease-

long intergenic non-coding RNA associations. IEEE

Transactions on Nanobioscience, 14 (2), 175-183.

Gonen, M. (2012). Predicting drug-target interactions

from chemical and genomic kernels using Bayesian

matrix factorization. Bioinformatics, 28 (18), 2304-

2310.

Gottlieb, A., Stein, G. Y., Ruppin, E., et al. (2011).

PREDICT: a method for inferring novel drug

indications with application to personalized medicine.

Molecular Systems Biology, 7, 496.

Hao, M., Bryant, S. H., & Wang, Y. (2017). Predicting

drug-target interactions by dual-network integrated

logistic matrix factorization. Scientific Reports, 7,

40376.

Hodos, R. A., Kidd, B. A., Shameer, K., et al. (2016). In

silico methods for drug repurposing and

pharmacology. Wiley Interdisciplinary Reviews-

Systems Biology and Medicine, 8 (3), 186-210.

Hunter, S., Apweiler, R., Attwood, T. K., et al. (2009).

InterPro: the integrative protein signature database.

Nucleic Acids Research, 37 (Database issue), D211-

215.

Jacob, L., & Vert, J. P. (2008). Protein-ligand interaction

prediction: an improved chemogenomics approach.

Bioinformatics, 24 (19), 2149-2156.

Kim, S., Chen, J., Cheng, T., et al. (2021). PubChem in

2021: new data content and improved web interfaces.

Nucleic Acids Research, 49 (D1), D1388-D1395.

Kohler, S., Bauer, S., Horn, D., et al. (2008). Walking the

interactome for prioritization of candidate disease

genes. American Journal of Human Genetics, 82 (4),

949-958.

Kuang, Q., Wang, M., Li, R., et al. (2014). A systematic

investigation of computation models for predicting

Adverse Drug Reactions (ADRs). PLoS One, 9 (9),

e105889.

Kuang, Q. F., Li, Y. Z., Wu, Y. M., et al. (2017). A kernel

matrix dimension reduction method for predicting

drug-target interaction. Chemometrics and Intelligent

Laboratory Systems, 162, 104-110.

Li, Y., & Patra, J. C. (2010). Genome-wide inferring gene-

phenotype relationship by walking on the

heterogeneous network. Bioinformatics, 26 (9), 1219-

1224.

Liu, Y., Wu, M., Miao, C., et al. (2016). Neighborhood

Regularized Logistic Matrix Factorization for Drug-

Target Interaction Prediction. PLoS Computational

Biology, 12 (2), e1004760.

Lotfi Shahreza, M., Ghadiri, N., & Green, J. R. (2019).

Heter-LP: A Heterogeneous Label Propagation

Method for Drug Repositioning. Methods in

Molecular Biology, 1903, 291-316.

Lotfi Shahreza, M., Ghadiri, N., Mousavi, S. R., et al.

(2017). Heter-LP: A heterogeneous label propagation

algorithm and its application in drug repositioning.

Journal of Biomedical Informatics, 68, 167-183.

Lotfi Shahreza, M., Ghadiri, N., Mousavi, S. R., et al.

(2018). A review of network-based approaches to drug

repositioning. Brief Bioinform, 19 (5), 878-892.

Luo, Y., Zhao, X., Zhou, J., et al. (2017). A network

integration approach for drug-target interaction

prediction and computational drug repositioning from

heterogeneous information. Nature Communications,

8 (1), 573.

Maleki, E. F., Ghadiri, N., Shahreza, M. L., et al. (2020).

DHLP 1&2: Giraph based distributed label

propagation algorithms on heterogeneous drug-related

networks. Expert Systems with Applications, 159.

Mei, J. P., Kwoh, C. K., Yang, P., et al. (2013). Drug-

target interaction prediction by learning from local

information and neighbors. Bioinformatics, 29 (2),

238-245.

Nagamine, N., Shirakawa, T., Minato, Y., et al. (2009).

Integrating statistical predictions and experimental

verifications for enhancing protein-chemical

interaction predictions in virtual screening. PLoS

Computational Biology, 5 (6), e1000397.

Nascimento, A. C., Prudencio, R. B., & Costa, I. G.

(2016). A multiple kernel learning algorithm for drug-

target interaction prediction. BMC Bioinformatics, 17,

46.

Paul, S. M., Mytelka, D. S., Dunwiddie, C. T., et al.

(2010). How to improve R&D productivity: the

pharmaceutical industry's grand challenge. Nature

Reviews Drug Discovery, 9 (3), 203-214.

Pliakos, K., & Vens, C. (2020). Drug-target interaction

prediction with tree-ensemble learning and output

space reconstruction. BMC Bioinformatics, 21 (1), 49.

Sachdev, K., & Gupta, M. K. (2019). A comprehensive

review of feature based methods for drug target

interaction prediction. Journal of Biomedical

Informatics, 93, 103159.

Saint-Antoine, M. M., & Singh, A. (2020). Network

inference in systems biology: recent developments,

challenges, and applications. Current Opinion

Biotechnology, 63, 89-98.

ICHIH 2022 - International Conference on Health Big Data and Intelligent Healthcare

Sing, T., Sander, O., Beerenwinkel, N., et al. (2005).

ROCR: visualizing classifier performance in R.

Bioinformatics, 21 (20), 3940-3941.

Sophic. (2012).

http://www.sophicalliance.com/documents/sophicdocs

/White%20Paper%20Update%201-27-

11/The%20Druggable%20Genome012511.pdf.

van Laarhoven, T., & Marchiori, E. (2013). Predicting

Drug-Target Interactions for New Drug Compounds

Using a Weighted Nearest Neighbor Profile. PLoS

One, 8 (6), e66952.

Wang, W., Yang, S., & Li, J. (2013). Drug target

predictions based on heterogeneous graph inference.

Pacific Symposium on Biocomputing, 53-64.

Wang, W., Yang, S., Zhang, X., et al. (2014). Drug

repositioning by integrating target information through

a heterogeneous network model. Bioinformatics, 30

(20), 2923-2930.

Wu, Z., Cheng, F., Li, J., et al. (2017). SDTNBI: an

integrated network and chemoinformatics tool for

systematic prediction of drug-target interactions and

drug repositioning. Briefings in Bioinformatics, 18 (2),

333-347.

Wu, Z., Li, W., Liu, G., et al. (2018). Network-Based

Methods for Prediction of Drug-Target Interactions.

Front Pharmacol, 9, 1134.

Xia, L. Y., Yang, Z. Y., Zhang, H., et al. (2019). Improved

Prediction of Drug-Target Interactions Using Self-

Paced Learning with Collaborative Matrix

Factorization. Journal of Chemical Information and

Modeling, 59 (7), 3340-3351.

Xia, Z., Wu, L. Y., Zhou, X., et al. (2010). Semi-

supervised drug-protein interaction prediction from

heterogeneous biological spaces. BMC Systems

Biology, 4 Suppl 2, S6.

Yamanishi, Y., Araki, M., Gutteridge, A., et al. (2008).

Prediction of drug-target interaction networks from the

integration of chemical and genomic spaces.

Bioinformatics, 24 (13), i232-240.

Yan, X. Y., Zhang, S. W., & Zhang, S. Y. (2016).

Prediction of drug-target interaction by label

propagation with mutual interaction information

derived from heterogeneous network. Mol Biosyst, 12

(2), 520-531.

Zhao, Q., Yu, H., Ji, M., et al. (2019). Computational

Model Development of Drug-Target Interaction

Prediction: A Review. Curr Protein Pept Sci, 20 (6),

492-494.

Networks and Algorithms in Heterogeneous Network-based Methods for Drug-target Interaction Prediction: A Survey and Comparison