Mathematical Biosciences 262 (2015) 147156

Contents lists available at ScienceDirect

Mathematical Biosciences
journal homepage: www.elsevier.com/locate/mbs

Bioinformatics in protein kinases regulatory network and drug discovery

Qingfeng Chen a,b,, Haiqiong Luo c , Chengqi Zhang d , Yi-Ping Phoebe Chen e,
a
School of Computer, Electronic and Information, Guangxi University, Nanning, 530004, China
b
State Key Laboratory for Conservation and Utilization of Subtropical Agro-bioresources, Guangxi University, China
c
School of Public Health, Guangxi Medical University, Nanning, 530021, China
d
Centre for Quantum Computation & Intelligent Systems, University of Technology, Sydney P.O. Box 123, Broadway, NSW 2007, Australia
e
Department of Computer Science and Computer Engineering, La Trobe University, Vic 3086, Australia

a r t i c l e i n f o a b s t r a c t

Article history: Protein kinases have been implicated in a number of diseases, where kinases participate many aspects
Received 18 July 2014 that control cell growth, movement and death. The deregulated kinase activities and the knowledge of
Revised 16 January 2015
these disorders are of great clinical interest of drug discovery. The most critical issue is the development
Accepted 22 January 2015
of safe and ecient disease diagnosis and treatment for less cost and in less time. It is critical to develop
Available online 2 February 2015
innovative approaches that aim at the root cause of a disease, not just its symptoms. Bioinformatics including
Keywords: genetic, genomic, mathematics and computational technologies, has become the most promising option for
Bioinformatics effective drug discovery, and has showed its potential in early stage of drug-target identication and target
Data mining validation. It is essential that these aspects are understood and integrated into new methods used in drug
Protein kinase discovery for diseases arisen from deregulated kinase activity. This article reviews bioinformatics techniques
Drug for protein kinase data management and analysis, kinase pathways and drug targets and describes their
Pathway potential application in pharma ceutical industry.
Disease
2015 Elsevier Inc. All rights reserved.

1. Introduction Traditional approach to drug discovery depends on trial-and-error

Protein kinases are viewed as the second most important group
of drug targets after G-protein-coupled receptors [22]. There are sev-
eral groups of protein kinases, and each group is then classied into
families or subfamilies [42]. Protein kinases are clinically relevant and
abnormal kinase activity is a frequent cause of a number of human dis-
eases. Nearly 400 human diseases have been reported to be connected
to protein kinases, such as cancer [74,87], cardiovascular [73,100],
neurological disorders [40,77], diabetes [76,83], rheumatoid arthri-
tis [84,101], and asthma [39,95,99]. Kinase activity is highly regulated
by phosphorylation, by combining activator proteins or inhibitor pro-
teins [65], or by changing their cellular location. The statistic data
indicate that nearly 2% of human genes, including 500 protein kinase
genes, are contained in the human genome [64]. Kinase activity has a
signicant effect on up to 30% of all human proteins. Thus, the inves-
tigation of features of kinase activity is an attractive therapeutic and
pharmaceutical strategy for drug design and the treatment of human
diseases.
Corresponding author. Tel.: +61 394796768.
E-mail addresses: qingfeng@gxu.edu.cn (Q. Chen), hqluo@163.com (H. Luo),
chengqi.zhang@uts.edu.au (C. Zhang), phoebe.chen@latrobe.edu.au (Y.-P.P. Chen).
of new chemical entities on cultured cells or animals, and matching
the apparent effects to treatments. Ligand-based drug design (indirect
drug design) and structure-based drug design (direct drug design) are
two major types of drug design [43]. However, the drug developed
from traditional methods might not be appropriate for all patients.
Some patients may be at risk of suffering serious side effects from new
drug. Further, traditional methods that largely depend on organism
level experimentation are time consuming and high cost.
Bringing a new drug to market from scratch typically takes
15 years and costs about $500 million. The pharmaceutical industry
are constantly searching for a better understanding of fundamental
disease mechanisms, tools for early diagnosis and even pre-diagnosis
disease [52]. The successful sequencing of the genomes of human and
other organisms in the past few years has opened the way to an en-
tirely new approach to drug design. A wealth of information on the
ingredients of patients at the genetic level is available due to the us-
age of bioinformatics. A number of algorithms and tools from data
mining, machine learning, articial intelligence, statistics have been
successfully used for gene identication and classication, secondary
structure prediction and function annotation [25,32]. In particular,
some of them have been applied in complementing disease diagnosis
and treatment of illnesses [77]. They will surely to play an essential
role in drug target discovery.

http://dx.doi.org/10.1016/j.mbs.2015.01.010
0025-5564/ 2015 Elsevier Inc. All rights reserved.
148 Q. Chen et al. / Mathematical Biosciences 262 (2015) 147156
Recent studies have shown that the structural variations [36] of
genome are implicated in a number of diseases and medical con-
ditions, ranging from genetic disorders to cancer, and are seen as
increasingly important in pharmaceutical research and development
and medical diagnostics [69]. The genome approach offers a blueprint
for ecient and personalized drug development. The tremendous
genome data make computational biology a central aspect for devel-
oping more advanced and highly customized therapies [31]. In addi-
tion, the technology may increase eciency and effectiveness of tests
for diagnosis of disease and patient-specic risk factors, and tools for
identication of individual patients likely to suffer from side effects
from taking certain drugs.
Many valuable genome data are generated owing to high through-
out biological techniques. Their management becomes a critical issue.
This may include gene data collection, gene annotation, structure
data modeling, standardization and normalization of data, database
establishment, and so forth. Further, the communication between
heterogenous biological database would be another key issue and
often requires data integration by removing inconsistent and noisy
data [16].
There have been an explosion of knowledge about signal transduc-
tion pathways, impacting virtually all areas of biology and medicine.
Protein kinases are key regulators of cell function that constitute one
of the largest and most functionally diverse gene families [49]. Fur-
ther, many diseases, such as cancer, diabetes and neurodegeneration,
indicate underlying gene correlations to the disease phenotype. The
study of in-depth knowledge of kinase pathways and possible role
to disease state is a big challenge [56]. There is still a long way to
go by combining the molecular biology, biochemistry, genetics and
bioinformatics for modern drug development [15].
Many protein kinase databases have been created to explore the
genomics, function and evolution of protein kinases. However, the
relevant data analysis and knowledge extraction for modern drug
discovery have been underdeveloped. Obviously, it is impractical to
handle this arduous and challenging problem by just relying on tradi-
tional biological experiments. In this regards, bioinformatics [53] that
include aspects of computer science, mathematics and molecular bi-
ology has become integral to process in that eld.
Bioinformatics has been widely applied to investigate the regu-
latory mechanisms of protein kinase, including their structural and
functional features [15]. Further, some researchers attempt to build
human protein kinase gene family and repository, by which to iden-
tify kinases that have a high probability of impacting human disease
based on data analysis [93]. This is able to discover useful information
from existing valuable data resources and greatly benet to the phar-
maceutical industry in the aspect of increasing accuracy and saving
cost. Although there have been many successful cases of bioinfor-
matics application in kinases and drug discovery. The bioinformatics
is still easily underestimated in both its cruciality and its resource
requirement [82]. This article aims to provide a literature review for
recent application and development of bioinformatics, protein kinase
regulatory network and drug discovery.
2. Protein kinase and diseases
Consistent with the complex role of the post-translational mod-
ication in the cell, protein kinases can be regulated by activator
proteins, inhibitor proteins, ligand binding to regulatory subunits,
cofactors, and phosphorylation by other proteins or by themselves
(autophosphorylation) [46,48]. Edmond H. Fischer and Edwin G. Krebs
were awarded the 1992 Nobel Prize in Physiology and Medicine for
discovering reversible protein phosphorylation as a biological regu-
latory mechanism.
Protein kinases have been viewed as a very attractive target class
for therapeutic interventions in many disease states such as can-
cer, diabetes, obesity, autoimmune disorders, inammation, vascular
diseases, and arthritis owing to their families key function in signal
transduction for all organisms. In this regard, protein kinases rep-
resent as much as 30% of all protein targets under investigation by
pharmaceutical companies. Protein kinases are novel and excellent
drug targets of post genomic era. Recent successful launches of drugs
with kinase inhibition as the mode of action demonstrate the ability
to deliver kinase inhibitors as drugs with the appropriate selectivity,
potency, and pharmacokinetic properties [22,96].
Bioinformatics is widely applied in identifying kinase-disease as-
sociations [81]. Ingenuity Pathway Analysis (IPA) software (Ingenuity
Systems, www.ingenuity.com) was used to construct sub-networks
signicantly associated with OSA (obstructive sleep apnea) [11]. The
results indicate a novel association of phosphoinositide 3-kinase,
the STAT family of proteins and its related pathways with OSA.
IPA, MetaCore (http://www.genego.com/metacore.php), sigPathway
algorithm (http://watson.nci.nih.gov/bioc_mirror/packages/2.3/bioc/
html/sigPathway.html) were carried out to understand the basis for
drug ecacy in the mouse model, by which to map similarities in
mTOR pathways in human lupus nephritis [72]. Gene set enrichment
analysis (GSEA) version 2.0 [89] identied biological pathways as-
sociated with resistance for each chemotherapy agent tested. Based
on the rank-ordered gene list provided by GSEA, the top genes up-
regulated and down-regulated for docetaxel resistance were ana-
lyzed using the connectivity map (cmap) to link genes associated
with a phenotype with potential therapeutic agents, such as the as-
sociation between phosphatidylinositol 3-kinase/AKT and docetaxel
resistance.
The kinase-disease associations can be seen at http://www.
cellsignal.com/reference/kinase_disease.html. It provides the infor-
mation of kinases, including their groups, disease types and molecular
basis. By using Hanks classication scheme [41,42], the human pro-
tein kinases can be clustered into groups, families, subfamilies on the
basis of the amino acid sequence similarity of their catalytic domains.
The reported diseases mainly consist of cancer, diabetes, cardiovascu-
lar, behavior, cardiopulmonary, neurodegeneration, vision, cognition,
hypertension, inammation. Table 1 presents a summary for kinase-
disease associations. It is observed that a disease type can be related to
multiple kinase groups, and several diseases can arise from a common
set of kinase group. As a result, the investigation of their associations
is useful to create regulatory pathways for drug discovery. For exam-
ple, AMPK (AMP-activated protein kinase) may have a regulatory role
in metabolism [38] and the therapeutic value of activating AMPK in
diabetes or metabolic syndrome is described in [20].
Abnormal protein phosphorylation has been proved to be a pri-
mary cause of disease. There has been increasingly growth of in-
terest in developing activate kinase inhibitors. There have been a
number of kinase-targeted drugs approved by experiments. Despite
some unsuccessful cases, some of them have been in the approval
of clinical trials. Protein kinases have now become the second most
Table 1
Summary of kinase-disease associations.
Disease types Kinase group
Cancer AGC, atypical, CAMK, CK1, CMGC, RGC, TK, TKL, STE,
Development AGC, atypical, CMGC, RGC, STE, TK, TKL
Diabetes AGC, CMGC, TK
Cardiovascular AGC, CAMK, CMGC, TKL
Behavior CK1, TKL
Hypertension AGC, CAMK, RGC
Neurodegeneration AGC, CAMK, CMGC, CK1
Inammation CMGC, STE, TKL
Vision AGC, RGC, TK
Epilepsy CAMK, TK
Cognition AGC, CMGC, STE, TKL
Immunity AGC, TK
Reproduction AGC, TKL
 Q. Chen et al. / Mathematical Biosciences 262 (2015) 147156
important group of drug targets since G-protein-coupled receptors
[22]. The target kinases for the approved drugs can be seen at http:
//en.wikipedia.org/wiki/Protein_kinase_inhibitor, in which some ex-
amples of kinase inhibitors in clinical use/trails and a comparison of
available agents used as human medicines are presented.
The rst oncogene is shown to be a protein kinase in 1978 [24].
It provided the rst connections between abnormal protein phos-
phorylation and disease. In the early 1980s, The rst protein-kinase
inhibitors, naphthalene sulfonamides, were introduced by Hiroyoshi,
from which several compounds were developed as inhibitors of pro-
tein kinases. Although many other compounds that were developed
subsequently are of relatively low potency in inhibiting protein ki-
nases, HA1077, also known as AT877 is progressed to human clin-
ical use in the early 1990s and was approved in Japan in 1995 for
the treatment of cerebral vasospasm. At the beginning, all inhibitors
are ATP competitive. According to the three-dimensional structure of
protein kinase, it was then proved to be dicult to develop protein-
kinase inhibitors with the requisite potency and specicity since the
residues that were involved in binding ATP were conserved from
kinase to kinase. In 1994, the rst nanomolar inhibitors of recep-
tor protein tyrosine kinases were developed and cytokine-synthesis
anti-inammatory drugs are indicated to inhibit p38 MAPK. Gleevec
(CGP57148, STI-571) was in human clinical trials for the treatment
of CML in 1996. In particular, just a small part of protein-kinase in-
hibitors can be used as drugs for the sake of toxicity, pharmacology
or solubility, such as SB203580. However, they could be very useful
research reagents.
The AMP-activated protein kinase (AMPK) system is a regulator of
energy balance at both the cellular and whole-body levels that, once
activated by low energy status, effects a switch from ATP-consuming
anabolic pathways to ATP-producing catabolic pathways. It now ap-
pears to be the major target for two existing classes of drug used
to treat type 2 diabetes, i.e., the biguanides and thiazolidinediones.
However, in both cases these activate AMPK indirectly, and an in-
teresting question concerns whether a drug that directly activated
AMPK would retain the therapeutic benets of the existing drugs
while eliminating unwanted side effects. AMPK activators also now
have potential as anticancer drugs.
The rapid development of bioinformatics, such as high-throughput
techniques and biology big data analysis, provides a smarter way to
identify or predict inhibitors of the activity and activation of particular
protein kinases in terms of a detailed understanding of their catalytic
and regulatory characteristic patterns, and is likely to play more and
more important roles in clinical use.
A number of kinase-targeted drugs have been developed in the
past few years. Although most were approved by FDA, some are
still in clinical trials, such as Fostamatinib by Rigel Pharmaceuticals,
Lenvatinib by Eisai Co. In particular, several inhibitors are proved to
be unsuccessful, and are possibly abandoned or withdrawn, such as
Mubritinib by Takeda, and Vandetanib by AstraZeneca . There are
27 FDA approved protein kinase inhibitors by April 2014.
Table 2
Resources of information about kinases.
Data source
KinMutbase (http://bioinf.uta./KinMutBase/)
Kinasource (http://www.kinasource.co.uk/Database/substrates.html)
Kinase.com (http://kinase.com/)
Kinasedb (http://kinasedb.ontology.ims.u-tokyo.ac.jp:8081/)
Kinweb (http://www.itb.cnr.it/kinweb/)
ProQinase (http://www.proqinase.com/content/view/82)
KSD (http://sequoia.ucsf.edu/ksd/)
149
Protein kinases have been viewed as the most important class of
drug target of cancer [23] in pharmaceutical industry. In the past
decade, 20 kinase-targeted drugs have been approved for clinical
use, and hundreds more are undergoing clinical trials. Further, the
recent approval of the rst protein kinase inhibitors for the treat-
ment of inammatory diseases, in combination with an enhanced
understanding of the signaling networks to control immune system,
could be a surge of interest in this area in the future. For exam-
ple, there have been efforts for investigating the connections be-
tween kinase and signalling networks, such as targeting protein ki-
nases in the MyD88 signaling network for the development of drugs
to treat chronic inammatory and autoimmune diseases, the pro-
tein kinases SPAK/OSR1 and LRRK2 for the treatment of hyperten-
sion and Parkinsons disease, and the progress in developing LRRK2
inhibitors.
3. Bioinformatics and protein kinase
3.1. Kinase data
Due to the application of advanced biological techniques, there
has been an increasing growth of diverse protein kinase data. The
tremendous data is not only valuable resource for deep understand-
ing of kinase pathways and is a big challenge for further data analysis
in a systematic way. A number of kinase databases have been estab-
lished for storage, management and other purposes. Table 2 presents
a collection of data sources of protein kinases.
KKB (kinase knowledgebase) is Eidogen-Sertantys database of
kinase structureactivity and chemical synthesis data. Eidogen-
Sertantys knowledgebase provides high quality training sets for com-
putational scientists to build predictive QSAR models. This is designed
to support medicinal chemists during all project stages of drug dis-
covery, and aims to ensure that the clients receive all the relevant
information around their targets and anti-targets.
KinBase explores the functions, evolution and diversity of protein
kinases, the key controllers of cell behavior. It focuses on the kinome,
the full complement of protein kinases in any sequenced genome.
The kinome of an organism represents the set of protein kinases in
its genome [50]. The term was rst used by Gerard Manning and col-
leagues in their papers analyzing the 518 human protein kinases [64]
and the evolution of protein kinases throughout eukaryotes. KinBase
holds information on over 3000 protein kinase genes found in the
genomes of human, and many other sequenced genomes. Users can
search the database by different gene names and accessions, or in
terms of the sequence based classication. It provides Blast analysis
of the human kinome by comparing mouse kinome to human kinome,
and carried out kinome analysis of several important organisms.
Kinweb includes a collection of protein kinases encoded in the
human genome. It provides: a comprehensive analysis of functional
domains of each gene product; a prediction of secondary and ter-
tiary structure motifs by using machine learning based programs; a
Contents
Information about kinases, gene and mutation in KinMutBase, and alignments
of sequences in KinMutBase
Kinase substrate
Kinome, the full complement of protein kinases in any sequenced genome
Classication of protein kinases and their functional conservation ortholog
tables among species, proteinprotein, proteingene, and proteincompound
interaction data, domain information, and structural information.
A collection of protein kinases encoded in the human genome
Protein Kinase Technology Platform for preclinical drug development of protein
kinase inhibitors in oncology and other therapeutic areas
A comprehensive compilation of aliases for each kinase
150 Q. Chen et al. / Mathematical Biosciences 262 (2015) 147156
collection of conserved sequence elements identied by comparative
analysis of human kinase genes and their murine counterparts, useful
to the identication of additional coding sequences. The users can
query the database by gene name or query the database by selecting
a classication group or a protein domain.
KSD (kinase sequence database) is a collection of protein kinase
sequences grouped into families by homology of their catalytic do-
mains. The aligned sequences are available in MS Excel format, as well
as in HTML. The current version of database features a total of 287
families, which contain 7128 protein kinases from 948 organisms.
KPD (kinase pathway database) is an integrated database in-
volving major completely sequenced eukaryotes. It also provides an
automatic pathway graphic image interface. The protein, gene, and
compound interactions are automatically extracted from abstracts for
all genes and proteins by natural-language processing (NLP). With this
database, pathways can be compared among species using data with
more than 47,000 protein interactions and protein kinase ortholog
tables.
Despite the valuable protein kinase data, their deep analysis are
insucient and their communication are underdeveloped. This pre-
vents us from obtaining a comprehensive understanding of the ki-
nase pathways and their potential usage in drug discovery. Thus, it
is critical to apply bioinformatics to deal with protein kinase data for
accurate and ecient drug design.
3.2. Phosphorylation
Protein phosphorylation is one type of post-translational mod-
ication of proteins, in which a serine(S)/threonine(T)/tyrosine(Y)
residue is phosphorylated by a protein kinase in combination with
phosphate group. Regulation of proteins in virtue of phosphoryla-
tion, namely phosphoregulation, has been viewed as one of the most
widely used modes responsible for activating, deactivating, or gives
rise to a change in protein function. Phosphorylation has been proved
to play central roles in regulating the subcellular location and activity
of the transcriptional repressor [30].
Kinases have been recognized as a metabolic master switch regu-
lating the majority of cellular pathways. A protein kinase alters other
proteins by chemically adding phosphate groups to them. Phosphory-
lation usually leads to a functional alteration of the target protein by
changing enzyme activity, cellular location, or association with other
proteins [9,12]. They regulate many aspects that control cell growth,
movement and death. Thus, it is critical to explore the correlations
between phosphorylation sites and protein kinases, especially the
distribution properties of amino acids around the phosphorylation
site of a specic kinase group.
In recent years, large-scale phosphorylation experiments were
performed to investigate the characteristic patterns of phosphopep-
tide. A large number of data phosphorylation sites are accumulated
in various database, such as Phospho.ELM and PhophoSitePlus. There
have been considerable efforts for phosphorylation prediction. Com-
putational methods for prediction of eukaryotic phosphorylation sites
was recently reviewed by Trost and Kusalik [94], including the ma-
chine learning technique used for handling sequence information,
structural information. This review summarizes, classies and com-
pares the computational techniques for phosphorylation site predic-
tion, and offers an overview of the challenges that are faced when de-
signing predictors and how they have been addressed. This aids users
to choose an appropriate phosphorylation site predictor for their spe-
cic or in general biological application, or to attempt to extend or
adapt existing techniques in the future.
The reversible phosphorylation of proteins is in responsible for
almost all aspects of cell life, while abnormal phosphorylation is
a cause or consequence of many diseases [21]. Mutations in par-
ticular protein kinases and phosphatases lead to a number of dis-
orders, and many naturally occurring toxins and pathogens exert
their effects by changing the phosphorylation states of intracellu-
lar proteins. There are about 518 protein kinases genes, constitut-
ing about 2% of all human genes [64]. It is estimated that one third
of the proteins can be phosphorylated and about half of the kinase
groups is connected to diseases like cancer [97]. Some specic in-
hibitors of protein kinases have been reported for the treatment of
cancer and chronic inammatory diseases [37]. For example, pro-
tein kinase (GSK3) may facilitate the development of drugs to treat
diabetes [86].
A variety of methods have been developed to explore phosphory-
lation. They focus on applying biology experiments to predict phos-
phorylation sites, such as mass spectrometry [63]. However, tradi-
tional experimental methods not only are high cost but also are easily
affected by many other reactions. Further, most discovered phospho-
rylation sites do not contain kinase information at all. For example,
there are less than 12% records in database Phospho.ELM [29] are
annotated by kinase groups. Most phosphopeptides do not contain
kinase information. A number of computational methods have been
proposed to deal with the increasingly growth of phosphorylated pro-
tein data, such as the KinasePhos2.0 using SVM [97], Netphosk using
neural network [8], PPSP using statistic, and PredPhospho based on
support vector machine [55]. Unfortunately, these algorithms aim to
identify phosphorylation sites, and are only able to deal with short
peptide sequence around phosphorylation sites. This may result in
substantial loss of information. Moreover, some useless information
may be introduced by just increasing sequence fragment length. This
can give rise to unexpected impact on the predicted results.
Protein kinases not only combine the adjacent domain to phospho-
rylation site of protein substrate, but also attempt to bind sequence
from remote domain, such as docking site [19]. In addition, there
also exist some known PBD (phosphopeptide-binding domains) with
relevance to combination of phosphopeptide [90]. Although docking
sites and PDB are a little far from phosphorylation site, they are im-
portant for its identication since both show more or less sequence
conservation. Thus, it is necessary to take into account the remote
domains to nd interesting featured patterns of phosphorylation
sites.
3.3. Mathematical models of kinase activity
Mathematical model is one of the most important methods to
depict integrated qualitative, dynamic and topological behavior or
activity of intracellular signal network, including diverse relation-
ships between involved components. There are many computational
modeling techniques applied for different purposes, such as Bayesian
network, HMM (hidden Markov model), neural network etc. Mathe-
matical models can be presented as many forms, including dynamic
systems, differentiate equation, statistical model, or logical models.
Although a model is initially for the topological representation of
its components and their correlations, the description in the model
of the biological systems dynamic behavior enables its predictive
power. The most important thing is to uncover and interpret those
unexpected behavior encoded in the design of biological networks. A
collection of models of the MAPK pathway are discussed in [57] from
different facets of emphasis, such as dual phosphorylations, feedback
loop. In general, a mathematical model usually represents a system
as a set of variables and a set of equations that establish relationships
between the variables.
Kinetic equations are applied to describe all molecular interactions
in [80]. The dissociation reaction is written as:
k+ kon
A + B A B C (1)
k koff
where k+ and kon represent the diffusion-dependent movement to-
ward the A B complex, and the following reaction, respectively.
 Q. Chen et al. / Mathematical Biosciences 262 (2015) 147156
Suppose kf and kr are the overall forward rate constant and overall
reverse rate constant, respectively. They are dened as
kon k+
kf = (2)
kon + k+
koff k
kr = (3)
koff + k
Kinetic parameter and state variables are two primary compo-
nents of the mathematical model and present the state of a system in
a certain time, such as the involved molecules of a specic compound.
As indicated by the authors, the former is derived from literature, and
includes MichaelisMenten constants, turnover numbers, and rate
constants of association and dissociation. It is based on ordinary dif-
ferential equations (ODES) and comprises 94 state variables and 95
parameters. Another kinetic rate-based model [92] is proposed for
the MAP kinase pathway, which comprises a cytosolic subsystem and
a nuclear subsystem. In particular, proteomic data are used for in-
ferring kinetic parameters and developing more accurate signaling
pathway. Genetic algorithm is applied to estimate the parameters,
and a MATLAB toolbox is used to infer the unknown rate constants.
Macromolecular crowding has been proved to impact on the intra-
cellular signaling pathway. However, the understanding of how the
macromolecular crowding affects the overall reaction rate and pro-
cessivity of enzymes is still insucient. Mathematical models of re-
action kinetics are proposed to address the effect of thermodynamic
activity, viscosity and processivity in an environment with macro-
molecular crowding [3]. Further, they provide empirical validations
by employing in vitro ERK MAP kinase phosphorylation.
A mathematical theory [45] is developed for modeling linear
kinase-phosphatase cascades, as well as systems containing feed-
back interactions, crosstalk with other signaling pathways, and/or
scaffolding and G proteins. Three key questions must be answered for
any signal transduction system, including (1) the speed that the signal
arrives the expected destination, (2) the time the signal lasts, and (3)
the strength of the signal. The authors introduce three parameters to
answer these questions, including (1) the signal time i , which means
the average time to activate kinase i, (2) the signal duration i , which
is the average time during which the kinase i is activated, and (3) the
signal amplitude Si , which is the average concentration of activated
kinase i. A phosphorylation step is viewed as a reaction between the
phosphorylated form Xi1 of kinase i 1 within the pathway and the
nonphosphorylated form X i of a downstream kinase i. The phospho-
rylation rate for each reaction is expressed as
vp,i = i Xi1 Xi (4)
where i is the second order rate constant for phosphorylation by the
ith kinase.
The dephosphorylation rate of the ith kinase is dened as
vd,i = i Xi (5)
where i is the rate constant for dephosphorylation by the ith phos-
phatase.
The concentration of each activated kinase i (except the rst ac-
tivated kinase in the pathway ) is represented as a function of time,
namely Xi (t), and is given by differential equations:
dXi
= vp,i vd,i = i Xi1 Xi i Xi (6)
dt
Computational model has been widely applied to understand the
behavior of complex biological systems by offering useful informa-
tion. Although it also performs many provocative predictions that
were not experimentally tested, most of the data that biologists gen-
erate cannot be directly amenable to dynamic modelling. They are
often qualitative instead of quantitative, include sparse time series,
151
and present relative changes instead of changes in absolute concen-
trations. Without standardization of measurements, users have to
compare the data from different laboratories in a qualitative form.
These block the use of ordinary differential equations (ODEs) since it
requires absolute changes in concentrations and exact reaction rates
as input. As a result, people expects to move seamlessly between
different levels of abstraction that can describe individual reactions,
network modules and whole networks.
3.4. Kinase data analysis for drug discovery
The kinase data analysis includes different aspects, such as features
of kinase sequence and structures, kinase classication and identi-
cation. There are many methods or tools used to address kinase data.
It is uneasy to describe all in this article. We focus on introducing typ-
ical bioinformatics algorithms or tools with respect to protein kinase
and drug discovery.
Discovery of kinase pathways. To fully understand protein kinase
networks, it is critical to identify regulators and substrates of kinases,
especially for weakly expressed proteins. To our knowledge, the used
bioinformatics techniques can include:
identifying kinase docking site;
clustering and visualization of phosphorylation proles of regula-
tors;
mining associations between kinase subunits, and between sub-
units and stimuli;
others.
A hybrid computational search algorithm has been developed,
which integrates machine learning and expert knowledge to discover
kinase docking sites. This algorithm was used to search the human
genome for novel MAP kinase substrates and regulators focused on
the JNK family of MAP kinases [91].
Cluster 3.0 [28] is used to classify the phosphorylation proles of
regulators, which are constructed by quantifying response regulator
bands in each prole. Further, the prole is visualized using Java
Treeview [75]. It is applied to dissect the basis of phosphotransfer
specicity in two-component signaling pathways in [10].
Association rule mining is used to analyze the AMPK regulation
data derived from the published experimental results [15]. A number
of rules of interest are discovered from mining AMPK data, such as
A = {moderate intensity treadmill} B = {high expression of 2a ,
high expression of 2a }. They reveal numerous potential associations
between the states of subunit isoforms of AMPK, or between the
stimulus factor and the state of isoforms, many of which are useful
for drug design. Further, negative rule association [98], represented
as the forms of X Y, X Y or X Y can be applied for
investigating the potential inhibitive regulatory correlation between
the subunit isoforms of protein kinases, and the stimulus factors.
Advanced bioinformatics and systems biology tools assist in un-
locking network properties and modeling kinase pathways. Recent
examples are the development of a predictor for breast cancer prog-
nosis based on the modularity of protein interaction networks and
the identication of cancer-associated phosphorylation networks
through the combined alignment of conserved phosphorylation sites
and kinase-substrate networks [58]. Moreover, Fuzzy logic model is
applied to encode the dynamics of a complex intracellular signal-
ing network of protein kinases [2]. In the similar way, an ecient
framework based on hidden Markov models (HMMs) is presented for
nding homologous pathways in a network of interest [71].
Kinases in gene expression. A rational approach is presented
for identifying and ranking protein kinases that are likely respon-
sible for observed changes in gene expression [5]. By combining pro-
moter analysis, it can identify and rank candidate protein kinases
for knock-down, or other types of functional validations, based on
152 Q. Chen et al. / Mathematical Biosciences 262 (2015) 147156
genome-wide changes in gene expression. This describes how pro-
tein kinase candidate identication and ranking can be made robust
by cross-validation with phosphoproteomics data as well as through
a literature-based text-mining approach. Data integration can rapidly
advance drug target discovery and unraveling drug mechanisms of
action.
Inappropriate activation of AKT signaling is a relatively common
occurrence in human tumors, and can be caused by activation of
components of, or by loss or decreased activity of inhibitors of this
signaling pathway. Causal network modeling is a systematic compu-
tational analysis that identies upstream changes in gene regulation
that can serve as explanations for observed changes in gene expres-
sion [60,61].
Kinase inhibition. While system biology and personalized
medicine are becoming increasingly important, there is a urgent need
to map the inhibition prole of a compound on a large panel of tar-
gets by using both experimental and computational methods. This is
especially important for kinase inhibitors, given the high similarity
at the binding site level for the 518 kinases in the human genome. A
new method is proposed and validated to predict the inhibition map
of a compound by comparison of binding pockets [67].
Kinase inhibitors have potential for treatment of many diseases.
Currently there are a number of drugs approved or in develop-
ment that target protein kinases. More details can be seen at http:
//en.wikipedia.org/wiki/Protein_kinase_inhibitor and will not be re-
peated herein. However, current inhibitors interact with a broad va-
riety of kinases and interfere with multiple vital cellular processes,
which causes toxic effects. Bioinformatics approaches that can pre-
dict inhibitorkinase interactions from the chemical properties of the
inhibitors and the kinase macromolecules might aid in design of more
selective therapeutic agents, that show better ecacy and lower tox-
icity. Proteochemometric modelling is applied to correlate the prop-
erties of 317 wild-type and mutated kinases and 38 inhibitors to the
respective combinations interaction dissociation constant (K d ) [62].
A report of protein kinases as targets for inhibitor design in [65] is
presented as follows:
Receptor tyrosine kinases. Dysregulation of growth factor sig-
naling networks has been reported in multiple human cancers.
Binding of growth factors to extracellular domains of receptor ty-
rosine kinases activates the intracellular kinase domain. Based on
EGFR (epidermal growth factor receptor) and VEGFR (the vascular
endothelial growth factor receptor), different therapeutic agents
have been developed for anticancer drug development.
Nonreceptor tyrosine kinases. About one third of tyrosine kinases
are classed as nonreceptor tyrosine kinases.
Phosphatidylinositol 3-kinase (PI3K). Resistance to radiation
treatment in a number of cancers has been linked to activation
of the PI3K-AKT pathway, which suggests that inhibition of PI3K
to overcome resistance and to improve the ecacy of radiation
treatment is an attractive clinical goal [66].
Signal-transducing serine-threonine kinases. p38 selective in-
hibition could be a therapeutically useful target route to treatment
of a number of inammatory and autoimmune diseases [70].
Cyclin-dependent kinases (CDKs) and other cell cycle control
kinases. Nature has provided many useful CDK inhibitors that have
reached clinical trials. High-throughput screening of compound
libraries has also disocovered many CDK inhibitors.
Computer-aided efforts, including molecular dynamics (MD) sim-
ulations and anisotropic network model (ANM) normal mode analy-
sis, are used for generating potential ligand-bound conformers start-
ing from the apo state of p38 [6]. A structure modeling-based method
is developed for sequence and structure analysis, which is helpful
in identifying the ligand binding sites and molecular function of the
Leishmania specic mitogen-activated protein kinase [78]. SiteAlign
algorithm [79] is used to measure the similarity of druggable protein-
ligand binding sites to the ATP-binding site of Pim-1 kinase (PDB entry
1yhs with bound inhibitor staurosporine) [27].
Kinase-inhibitor proling panel is another important kinase-
specic topic. Many researchers proposed high-throughput prol-
ing studies for interaction between many protein kinases versus
many chemical compounds. Some of these huge data are publicly
accessed, such as moleculekinase interaction map for clinical ki-
nase inhibitors [33] and a systematic interaction map of validated
kinase inhibitors with Ser/Thr kinases [34]. A number of informatics
studies based on these data have been published, and are recently
summarized in [35]. It presents machine learning methods either for
classication (binds/does not bind) or regression on the measured
inhibition values. Variants of nave Bayesian classier, support vector
machine, decision tree, k-nearest neighbors were used to learn asso-
ciations between kinase residues and compound fragments. Another
class of works includes those aims to obtain a more accurate repre-
sentation of kinase binding sites by taking advantage of kinase 3D
structures.
4. Kinase genes and drug targets
Many kinase genes have been identied in the human genomes,
and other sequenced genomes. For example, KinBase holds infor-
mation on over 3000 protein kinase genes. Many databases provide
searching tools in terms of a variety of different gene names and ac-
cessions, or according to the sequence based classication. The chal-
lenge to bioinformatics is evolving from that of identifying long lists
of genes to that of determining short lists of the targets most likely to
play central roles in diseases.
A variety of gene identication algorithms have been developed
in drug discovery, including the identication of positions, secondary
structures and functions. Nevertheless, the discussion of these al-
gorithms are not the key point in this article. In contrast, this pa-
per focuses on describing the associations with respect to kinase
genes and drug targets. It is presented in terms of the following
aspects.
Kinase gene and diseases
Human kinome in drug discovery
4.1. Kinase genes and diseases
The gene family encoding protein kinases is the most commonly
mutated in human cancer. Moreover, mutated and activated protein
kinases have proved to be tractable targets for the development of
new anticancer therapies. At http://www.sanger.ac.uk/genetics/CGP/
Kinases/, it examined the full coding sequence of the protein kinase
genes (518 protein kinases, more than 1.3 Mb of DNA per sample)
in primary cancers and cancer cell lines. Targeting receptor protein
tyrosine kinases (RPTKs) as a cancer chemotherapy has continued
to become a compelling approach for a long time. Preclinical and
clinical data strongly support the involvement of specic RPTKs in
the formation and progression of a subset of solid and liquid tumors.
As mentioned above, there are various kinases that are related
to different diseases, such as cancer and diabetes. To explore the
functions, evolution and diversity of protein kinases, it is necessary
to transfer our focus to their relevant sequenced genome. This assists
us in obtaining a deep and comprehensive understanding of kinase
pathways with respect to drug discovery.
It has been commonly recognized that gene mutation, including
deletion, insertion and duplication, can result in diseases. No matter
which case occurs, the protein made by the gene may not function
properly. In other words, the mutation can alter the function of the re-
sulting protein. Thus, the investigation of disease relevant kinase gene
 Q. Chen et al. / Mathematical Biosciences 262 (2015) 147156
Table 3
An example of kinase-gene associations.
Gene Species Kinase classication
AKT1 Human AGC: Akt
AKT2 Human AGC: Akt
AKT3 Human AGC: Akt
BARK1 Human AGC: GRK: BARK
MAST1 Human AGC: MAST: MAST
families is critical. Meanwhile, their analysis including sequences and
structure is also a big challenge for us.
Table 3 presents a list of genes with respect to specied kinase
categories. Suppose one gene is found to be related to a kinase group,
or its families or subfamilies, we are able to link genes and diseases via
kinases. For example, in Table 1, AGC kinase group is a frequent cause
of many diseases. Kinase types can be viewed as the key, by which
a comprehensive regulatory networks for protein kinases, genes and
diseases can be generated.
The leader gene approach, a data mining method, for gene search
in a specic process is used to identify the common genomic path-
ways between periodontitis and type 2 diabetes [26]. ENDEAVOUR
software is applied to prioritize all genes of the whole genome
in relation to type 2 diabetes [88]. A regularized Bayesian inte-
gration system, HEFalMp (Human Experimental/Functional Mapper,
http://function.princeton.edu/hefalmp), aims to provide maps of
functional activity and interaction networks in over 200 areas of
human cellular biology. It allows users to interactively explore func-
tional maps integrating evidence from thousands of genomic experi-
ments, focusing as desired on specic genes, processes, or diseases of
interest [54].
4.2. Kinase gene identication
A number of algorithms and tools are developed for gene identi-
cation. As described in [59], they can include software tools for (1) ab
initio gene prediction, (2) splicing site prediction, and (3) homology-
based prediction. ab initio gene prediction relies on the own informa-
tion in the DNA sequence, such as promoter, intro or exon, and use
statistical parameter to predict genes. In contract, homology-based
prediction depends primarily on nding homologous sequences in
other genomes and/or in public databases using BLAST, or Smith
Waterman algorithms. This method compares newly obtained se-
quence data from experiments with known gene information. It uses
sequence alignment to construct gene models and validates the pre-
dicted genes through similarity searches including sequence similar-
ity or structure similarity. Splicing site prediction is commonly used
as complement in the gene prediction tools.
Many software tools can be publicly accessed for ab ini-
tio gene prediction. The representative works include Gene-
Mark (http://exon.gatech.edu/GeneMark/), GENSCAN (http://genes.
mit.edu/GENSCAN.html), GeneBuilder (http://http://zeus2.itb.cnr.
it/~webgene/genebuilder.html/). GeneMark employed a non-
homogeneous Markov model to classify DNA regions into protein-
coding, non-coding, and non-coding but complementary to coding.
GENSCAN uses a complex probabilistic model of the gene structure
that is based on actual biological information about the properties of
transcriptional, translational, and splicing signals. GeneBuilder is an
integrated computing system for protein-coding gene prediction.
NNSplice (http://www.fruity.org/seq_tools/splice.html). Splice-
view (http://zeus2.itb.cnr.it/~webgene/wwwspliceview.html), Net-
Gene2 (http://www.cbs.dtu.dk/services/NetGene2/) are three typical
methods for splicing site prediction. NNSplice uses a decision tree
method called maximal dependence decomposition (MDD), and en-
hances it with Markov models that capture additional dependencies
among neighboring bases in a region around the splice site. Splice-
view is based on prediction of splice signals by classication approach.
153
Neural networks in combination with sequence similarity are trained
to recognize splice sites by NetGene2.
PipMaker (http://pipmaker.bx.psu.edu/pipmaker/) and Ge-
neWise http://www.ebi.ac.uk/Tools/Wise2/index.html) are tools for
homology-based prediction. The former compares two long DNA
sequences to identify conserved segments in terms of the alignment
engine called BlastZ and a scoring matrices. The latter was developed
from a combination of hidden Markov models to predict gene
structure using similar protein sequence.
Sequence similarity and structure similarity are widely used
strategies in gene identication for functional genomic. A variety of
approaches have been developed for measuring the similarities, such
as BLAST and FASTA for sequence similarity, DALI (distance alignment
matrix method used a Monte Carlo simulation) [47] for protein align-
ment and FOLDALIGN for RNA structural alignment [44]. Recent stud-
ies indicate that RNA structures perform many important regulatory,
structural, and catalytic roles in the cell. They have been found in most
organism and comprise functional domains within ribozymes, self-
splicing introns, ribonucleoprotein complexes, viral genomes, and
many other biological systems [1]. Association rule mining is applied
to identify hidden structurefunction patterns within RNA [13,14].
There are also a number of reported approaches and case studies
that specialize in kinase gene identication. Thirty-four CDPK genes
that are widespread in plant consist of a multigene family were iden-
tied by a genome-wide analysis of Arabidopsis CDPKs [17]. However,
only a limited number of rice CDPK family members have been char-
acterized. A database search was performed to identify CDPK family
genes in the rice genome using entire amino acid sequences from
previously identied rice CDPKs and the full-length rice cDNA clones
encoding CDPKs [4]. The genomic sequences were selected as can-
didates according to the score by the TBLASTN algorithm, and the
obtained candidate sequences were analyzed by the BLASTX program
and motif scanning, by which to sort out CDPKs from other related
protein kinases. By the above search, 29 rice CDPKs are discovered.
A set of host kinase genes required for inuenza virus replication
and the regulatory role of microRNAs were identied [7], in which a
small interfering RNA (siRNA) screen of 720 HPKs was performed. It
assists in understanding the role of HPKs (human protein kinases) and
the signaling networks that impact inuenza virus replication. Up-
regulation and down-regulation of transcript expression of induced
protein kinases by miRNAs and protein expression are applied by
performing an RNAi-based genetic screen.
Eukaryotic protein kinases (PKs) represent one of the largest pro-
tein superfamilies [18]. PKs are related by the presence of a con-
served kinase domain (or catalytic domain). Based on the substrate
specicity, PKs can be further divided into protein tyrosine kinases
(PTKs) and protein serine/threonine kinases (PSKs). Genomic DNA is
applied as template to perform PCR (polymerase chain reaction) to
screen for protein kinase (pk) genes for the round-spotted puffer-
sh Tetraodon uviatilis. Forty-one T. uviatilis pk genes encoding 7
receptor tyrosine kinases, 14 nonreceptor tyrosine kinases, 16 ser-
ine/threonine kinases, 1 dual kinase and 3 novel kinases have been
identied.
4.3. Challenges in drug discovery
The completion of Human Genome project had assembled a ge-
netic blueprint for human being. Genome science has accelerated
genetic research and revolutionized the diagnosis, prevention and
treatment of various diseases. Genetics has showed its signicance in
clinical medicine. It collects and studies medical histories and DNA
samples from family members. Genetic analysis of these samples as-
sists researchers in nding genes or patterns of genes that are different
among affected and unaffected family members and that may be re-
lated to the disease. Genetic testing aids in dening clinical indicators
of a hereditary predisposition to develop cancers.
154 Q. Chen et al. / Mathematical Biosciences 262 (2015) 147156
In recent years, there have been an explosive growth of dis-
ease genes such as oncogenes found due to the high throughput
screening. For example, Cancer Gene (https://cabig.nci.nih.gov/
inventory/data-resources/cancer-gene-index) Index is a collection
of records on 6955 human genes identied from the literature as
having an association with cancer. Diabetes Disease Portal (http:
//rgd.mcw.edu/) provides gene information for several diseases, such
as diabetes, obesity and metabolic diseases. It is observed that genes
and diseases might be cross-correlated. Bioinformatics is an ecient
way to validate those discovered targets and sort out short lists of the
targets most possibly to be critical in disease.
The identication of susceptibility genes for cancers allows testing
before symptoms become apparent. If a single gene is responsible,
testing during pregnancy or at any other appropriated time of life for
the specic cancer may predict a high risk or eliminate concern about
that particular cancer. For example, if a person is found to carry an
inherited mutation in one of the cancer mutation repair genes, he/she
could benet from annual xed examination. Thus, any pathological
changed would be detected and eliminated before they progress to a
potentially invasive cancer.
Microarray allow the gene expression proles of thousands of
genes to be measured and compared with each other in cells and
tissues between healthy and diseased states. Those genes that are
functionally associated usually share similar expression proles. To
ensure the accuracy, the microarray data is combined with pathways
that exist in the context of complex proteinprotein interaction net-
works. It is necessary to use gene ontologies (GO), which describes
gene products in terms of their associated biological processes, cel-
lular components and molecular functions in a species-independent
manner.
Most of the genetic disorders originate from a mutation in one
gene. One of the most dicult issues is to determine how genes con-
tribute to diseases that have a complex pattern of inheritance, such
as in the cases of diabetes, asthma and cancer. In other words, no
one gene is able to denitely say whether a person has a disease or
not. It is possible that several mutations occur before the disease is
evident, and many genes may each partially contribute to a persons
susceptibility to a disease; genes may also affect a persons reactions
to environmental factors. It is a big challenge to unravel these net-
works of events. It will be undoubtedly assisted by the availability of
the sequence of the human genome [51].
Kinase SARfari is an integrated chemogenomics database for
kinase, which can be accessed by https://www.ebi.ac.uk/chembl/
sarfari/kinasesarfari/. The system focuses on the protein kinase family
of drug targets. It combines chemical and biological data and provides
a platform that integrates kinase bioactivity data and links bioactiv-
ities to kinase sequence, structure, compounds and screening data.
It comes populated with diverse chemical and biological data re-
sources, which comprises all human protein kinase sequences and
a large number of model organism orthologs; protein kinase clini-
cal candidates and FDA-approved drugs. In particular, more than 700
3D structural domains from the protein data bank (PDB) with com-
plete binding-site focused structural superposition, more than 14,000
protein kinase compounds and more than 57,000 structureactivity
relationship (SAR) screening are included. This greatly increases pro-
ductivity and facilitate knowledge discovery.
Most kinase inhibitors under development as drugs act by directly
competing with ATP at the ATP-binding site of the kinase. There are
more than 500 protein kinases, and the ATP-binding site is highly con-
served among them. Therefore selectivity is an essential requirement
for clinically effective drugs, and understanding the structural char-
acteristics of ATP-binding sites is of crucial importance. The objective
of the present study was to elucidate the structural characteristics
of the adenosine-binding site of four major kinase groups, AGC (PKA,
PKG, and PKC families), CaMK (calcium/calmodulin-dependent pro-
tein kinases), CMGC (CDK, MAPK, GSK3, and CLK families), and TK
(tyrosine kinases). To do this, the kinases is classied into groups by
using feed-forward multilayer perceptron (MLP) neural networks and
structural, electronic, and hydrophobic descriptors of the amino acids
at the adenosine-binding site [68].
In general, tradition virtual drug screening has a limitation, namely
assuming a well-dened binding site. A ligand binding pathway anal-
ysis for kinase using long time molecular dynamics simulation was
performed by Shan et al. to identify target binding site [85]. Unlike
traditional methods, it does not demand for any prior knowledge of
the binding sites location. This may assist in the development of al-
losteric inhibitors that target previously undiscovered binding sites.
Further, due to identication of previously unknown binding sites,
molecular dynamics(MD) simulations of protein-ligand binding, may
greatly extend the applicability of computational techniques to drug
development.
5. Conclusions
Emerging varied diseases have been a big challenge to current
health care system. Many evidences indicate protein kinases are in-
volved in a number of diseases. Many diseases have their roots in
genes. Thus, the study of abnormalities, disorder or mutation of ki-
nase genes is a promising and ecient way for drug target discovery
because they can change the function of genes and have an effect on
relevant diseases. The discovery of protein kinase networks including
kinase types, diseases, inhibitors and kinome assist us in understand-
ing the root cause of diseases. Bioinformatics combing genetics and
genomics technologies has become a critical aspect of drug discov-
ery. Applying bioinformatics into target discovery and data manage-
ment can not only increase the eciency and accuracy of drug design
and generate personalized drug development. This article reviews
the techniques for protein kinase networks, including kinase data,
kinase inhibitors, kinase gene families, and their properties. Further,
the relevant algorithms, tools and the associations between kinome
and drug discovery are highlighted.
Acknowledgments
The work reported in this paper was partially supported by a
National Natural Science Foundation of China project 61363025 and
two key projects of Natural Science Foundation of Guangxi, Guangxi,
China 053006 and 019029.
References
[1] D.P. Aalberts, N.O. Hodas, Asymmetry in RNA pseudoknots: Observation and
theory, Nucleic Acids Res. 33 (7) (2005) 22102214.
[2] B.B. Aldridge, J. Saez-Rodriguez, J.L. Muhlich, P.K. Sorger, D.A. Lauffenburger,
Fuzzy logic analysis of kinase pathway crosstalk in TNF/EGF/INSULIN-induced
signaling, PLoS Comput. Biol. 5 (4) (2009) e1000340.
[3] K. Aoki, K. Takahashi, K. Kaizu, M. Matsuda, A quantitative model of ERK MAP
kinase phosphorylation in crowded media, Sci. Rep. 3 (2013) 1541.
[4] T. Asano, N. Tanaka, G. Yang, N. Hayashi, S. Komatsu, Genome-wide identication
of the rice calcium-dependent protein kinase and its closely related kinase gene
families: comprehensive analysis of the CDPKs gene family in rice, Plant Cell
Physiol. 46 (2) (2004) 356366.
[5] M. Avi, C.H. John, Protein kinase target discovery from genome-wide messenger
RNA expression proling, Mt. Sinai J. Med. 77 (4) (2010) 345349.
[6] A. Bakan, I. Bahar, Computational generation inhibitor-bound conformers of
p38 map kinase and comparison with experiments, Proc. Pac. Symp. Biocomput.
(2011) 181192.
[7] A. Bakre, L.E. Andersen, V. Meliopoulos, et al. Identication of host kinase genes
required for inuenza virus replication and the regulatory role of microRNAs,
PLOS One 8 (2013) e66796.
[8] N. Blom, T. Sicheritz-Ponten, R. Gupta, S. Gammeltoft, S. Brunak, Prediction of
post-translational glycosylation and phosphorylation of proteins from the amino
acid sequence, Proteomics 4 (6) (2004) 16331649.
[9] D. Bungard, B.J. Fuerth, P.Y. Zeng, B. Faubert, N.L. Mass, B. Viollet, D. Carling,
C.B. Thompson, R.G. Jones, S.L. Berger, Signaling kinase AMPK activates stress-
promoted transcription via histone H2B phosphorylation, Science 329 (5996)
(2010) 12011205.
 Q. Chen et al. / Mathematical Biosciences 262 (2015) 147156
[10] E.J. Capra, B.S. Perchuk, E.A. Lubin, O. Ashenberg, J.M. Skerker, et al. Systematic
dissection and trajectory-scanning mutagenesis of the molecular interface that
ensures specicity of two-component signaling pathways, PLoS Genet. 6 (11)
(2010) e1001220.
[11] E. Castiglioni, et al. Sequence variations in mitochondrial ferritin: distribution
in healthy controls and different types of patients, Genet. Test. Mol. Biomarkers
14 (6) (2010) 793796.
[12] Q.F. Chen, Y.P.P. Chen, Mining inhibition pathways for protein kinases on skeletal
muscle, IEEE Intell. Syst. 27 (5) (2012) 1926.
[13] Q.F. Chen, Y.P.P. Chen, Mining characteristic relations bind to RNA secondary
structures, IEEE Trans. Inform. Technol. Biomed. 14 (1) (2010) 1015.
[14] Q.F. Chen, Y.P.P. Chen, Discovery of structural and functional features in RNA
pseudoknots, IEEE Trans. Knowl. Data Eng. 21 (7) (2009) 974984.
[15] Q.F. Chen, Y.P.P. Chen, Mining frequent patterns for AMP-activated protein reg-
ulation on skeletal muscle, BMC Bioinform. 7 (2006) 394.
[16] C. QF, Y.P.P. Chen, C.Q. Zhang, Detecting inconsistency in biological molecular
databases using ontology, Data Min. Knowl. Discov. 15 (2) (2007) 275296.
[17] S.H. Cheng, M.R. Willmann, H.C. Chen, J. Sheen, Calcium signaling through pro-
tein kinases. The arabidopsis calcium-dependent protein kinase gene family,
Plant Physiol. 129 (2002) 469485.
[18] C.M. Chou, W.C. Lin, J.H. Leu, T.L. Su, C.K. Chou, C.J. Huang, Isolation and identica-
tion of novel protein kinase genes from the round-spotted puffersh (Tetraodon
uviatilis) genomic DNA, J. Biomed. Sci. 5 (2) (1998), 127134.
[19] A.J. Bardwell, E. Frankson, L. Bardwell, Selectivity of docking sites in MAPK ki-
nases, J. Biol. Chem. 284 (2009) 1316513173.
[20] M.P. Coghlan, D.M. Smith, Introduction to the kinases in diabetes biochemical
society focused meeting: are protein kinases good targets for antidiabetic drugs?
Biochem. Soc. Trans. 33 (2) (2005) 339342.
[21] P. Cohen, The role of protein phosphorylation in human health and disease, Eur.
J. Biochem. 268 (19) (2001) 50015010.
[22] P. Cohen, Protein kinases the major drug targets of the twenty-rst century?
Nat. Rev. Drug Discov. 1 (4) (2002) 309315.
[23] P. Cohen, D.R. Alessi, Kinase drug discoverywhats next in the eld? ACS Chem.
Biol. 8 (1) (2013) 96104.
[24] M.S. Collett, R.L. Erikson, Protein kinase activity associated with the avian sar-
coma virus src gene product, Proc. Natl. Acad. Sci. USA 75 (1978) 20212024.
[25] G. Cong, K.L. Tan, K.H. Anthony, T. Xin, X. Xu, Mining top-K covering rule groups
for gene expression data, in: Proceedings of the 2005 ACM SIGMOD International
Conference on Management of Data, 2005, pp. 670681.
[26] U. Covani, S. Marconcini, G. Derchi, A. Barone, L. Giacomelli, Relationship be-
tween human periodontitis and type 2 diabetes at a genomic level: a data-mining
study, J. Periodontol. 80 (8) (2009) 12651273.
[27] E. Defranchi, C. Schalon, M. Messa, F. Onofri, F. Benfenati, D. Rognan, Binding
of protein kinase inhibitors to synapsin I inferred from pair-wise binding site
similarity measurements, PLoS One 5 (8) (2010) e12214.
[28] M.J. de Hoon, S. Imoto, J. Nolan, S. Miyano, Open source clustering software,
Bioinformatics 20 (2004) 14531454.
[29] H. Dinkel, C. Chica, A. Via, C.M. Gould, L.J. Jensen, T.J. Gibson, F. Diella, Phos-
pho.ELM: a database of phosphorylation sitesupdate, Nucleic Acids Res. 39
(Database issue) (2011) D261D267.
[30] D. Domnguez, B. Montserrat-Sents, A. Virgs-Soler, et al. Phosphorylation reg-
ulates the subcellular location and activity of the snail transcriptional repressor,
Mol. Cell Biol. 23 (14) (2003) 50785089.
[31] J.T. Dudley, Y. Pouliot, R. Chen, A.A. Morgan, A.J. Butte, Translational bioinfor-
matics in the cloud: an affordable alternative, Genome Med. 2 (2010) 51.
[32] S.R. Eddy, Non-coding RNA genes and the modern RNA world, Nat. Rev. Genet.
2 (2001) 919929.
[33] M.A. Fabian, W.H. Biggs, D.K. Treiber, et al. A small molecule-kinase interaction
map for clinical kinase inhibitors, Nat. Biotechnol. 23 (3) (2005) 329336.
[34] O. Fedorov, B. Marsden, V. Pogacic, P. Rellos, S. Mller, A.N. Bullock, J. Schwaller,
M. Sundstrm, S. Knapp, A systematic interaction map of validated kinase in-
hibitors with Ser/Thr kinases, Proc. Natl. Acad. Sci. USA 104 (51) (2007) 20523
20528.
[35] F. Ferr, A. Palmeri, M. Helmer-Citterich, Computational methods for analysis
and inference of kinase/inhibitor relationships, Front. Genet. 5 (2014) 196.
[36] L. Feuk, A.R. Carson, S.W. Schererc, Structural variation in the human genome,
Nat. Rev. Genet. 7 (2006) 8597.
[37] J.G. Foster, M.D. Blunt, E. Carter, S.G. Ward, Inhibition of PI3K signaling spurs new
therapeutic opportunities in inammatory/autoimmune diseases and hemato-
logical malignancies, Pharmacol. Rev. 64 (4) (2012) 10271054.
[38] L.G.D. Fryer, D. Carling, AMP-activated protein kinase and the metabolic syn-
drome, Biochem. Soc. Trans. 33 (2) (2005) 362366.
[39] J.M. Goldberg, G. Manning, A. Liu, P. Fey, K.E. Pilcher, Y.J. Xu, J.L. Smith, The dic-
tyostelium kinomeanalysis of the protein kinases from a simple model organism,
PLoS Genet. 2 (3) (2006) 02910303.
[40] Z. Guo, S. Kozlov, M.F. Lavin, M.D. Person, T.T. Paull, ATM activation by oxidative
stress, Science 330 (6003) (2010) 517521.
[41] S. Hanks, A. Quinn, T. Hunter, The protein kinase family: conserved features and
deduced phylogeny of the catalytic domains, Science 241 (1988) 4252.
[42] S.K. Hanks, T. Hunter, Protein kinases 6. The eukaryotic protein kinase superfam-
ily: kinase (catalytic) domain structure and classication, FASEB J. 9 (8) (1995)
576596.
[43] H. Jhoti, A.R. Leach, Structure-Based Drug Discovery, Springer, (2007).
[44] J.H. Havgaard, R.B. Lyngso, G.D. Stormo, J. Gorodkin, Pairwise local structural
alignment of RNA sequences with sequence similarity less than 40%, Bioinfor-
matics 21 (9) (2005) 18151824.
155
[45] R. Heinrich, B.G. Neel, T.A. Rapoport, Mathematical models of protein kinase
signal transduction, Mol. Cell 9 (2002) 957970.
[46] S. Higashiyama, H. Iwabuki, C. Morimoto, M. Hieda, H. Inoue, N. Matsushita,
Membrane-anchored growth factors, the epidermal growth factor family: be-
yond receptor ligands, Cancer Sci. 99 (2) (2008) 214220.
[47] L. Holm, C. Sander, Mapping the protein universe, Science 273 (1996) 595603.
[48] http://www.abgent.com/docs/article_kinases.
[49] http://www.cellsignal.com/reference/kinase/index.html.
[50] http://kinase.com/.
[51] http://www.ncbi.nlm.nih.gov/books/NBK22183/.
[52] http://www.sciencemag.org/site/products/drugdiscnew.xhtml.
[53] L. Hunter (Ed.), Articial Intelligence and Molecular Biology, MIT Press, 1993.
[54] C. Huttenhower, E.M. Haley, M.A. Hibbs, V. Dumeaux, D.R. Barrett, H.A. Coller,
O.G. Troyanskaya, Exploring the human genome with functional maps, Genome
Res. 19 (6) (2009) 10931096.
[55] J.H. Kim, J. Lee, B. Oh, K. Kimm, I. Koh, Prediction of phosphorylation sites using
SVMs, Bioinformatics 20 (2004) 31793184.
[56] H. Kitano, A robustness-based approach to systems-oriented drug design, Nat.
Rev. Drug Discov. 6 (2007) 202210.
[57] W.1. Kolch, M. Calder, D. Gilbert, When kinases meet mathematics: the systems
biology of MAPK signalling, FEBS Lett. 579 (8) (2005) 18911895.
[58] W. Kolch, A. Pitt, Functional proteomics to dissect tyrosine kinase signalling
pathways in cancer, Nat. Rev. Cancer 10 (9) (2010) 618629.
[59] E.V. Koonin, M.Y. Galperin, Sequence-Evolution-Function: Computational Ap-
proaches in Comparative Genomics, Kluwer Academic, 2003.
[60] R. Kumar, S.J. Blakemore, C.E. Ellis, E.F. Petricoin, D. Pratt, M. Macoritto,
A.L. Matthews, J.J. Loureiro, K. Elliston, Causal reasoning identies mechanisms
of sensitivity for a novel AKT kinase inhibitor, GSK690693, BMC Genom. 11
(2010) 419.
[61] J. Lamb, E.D. Crawford, D. Peck, et al. The connectivity map: using gene-
expression signatures to connect small molecules, genes, and disease, Science
313 (2006) 19291935.
[62] M. Lapins, J.E.S. Wikberg, Kinome-wide interaction modelling using alignment-
based and alignment-independent approaches for kinase description and linear
and non-linear data analysis techniques, BMC Bioinform. 11 (2010) 339.
[63] J. Lin, Z. Xie, H. Zhu, J. Qian, Understanding protein phosphorylation on a systems
level, Brief. Funct. Genom. 9 (1) (2010) 3242.
[64] G. Manning, D.B. Whyte, R. Martinez, T. Hunter, S. Sudarsanam, The protein
kinase complement of the human genome, Science 298 (2002) 19121934.
[65] E.M. Noble Martin, A. Endicott Jane, N. Johnson Louise, Protein kinase inhibitors:
insights into drug design from structure, Science 303 (5665) (2004) 1800
180519.
[66] W.G. McKenna, R.J. Muschel, Genes chromosomes, Cancer 38 (2003) 330.
[67] F. Milletti, A. Vulpetti, Predicting polypharmacology by binding site similar-
ity: from kinases to the protein universe, J. Chem. Inf. Model 50 (2010) 1418
1431.
[68] T. Niwa, Elucidation of characteristic structural features of ligand binding sites
of protein kinases: a neural network approach, J. Chem. Inf. Model. 46 (5) (2006)
21582166.
[69] C.T. ODushlaine, R.J. Edwards, S.D. Park, D.C. Shields, Tandem repeat copy-
number variation in protein-coding regions of human genes, Genome Biol. 6
(2006) R69.
[70] G. Pearson, et al. Mitogen-activated protein (MAP) kinase pathways: regulation
and physiological functions, Endocr. Rev. 22 (2001) 153.
[71] X.N. Qian, S.H. SZE, B.J. Yoon, An ecient framework based on hidden markov
models (HMMS) that can be used for nding homologous pathways in a network
of interest, J. Comput. Biol. 16 (2) (2009) 145157.
[72] P.S. Reddy, H.M. Legault, J.P. Sypek, M.J. Collins, E. Goad, S.J. Goldman, W. Liu,
S. Murray, A.J. Dorner, M. OToole, Mapping similarities in mTOR pathway per-
turbations in mouse lupus nephritis models and human lupus nephritis, Arthritis
Res. Ther. 10 (6) (2008) R127.
[73] B.A. Rose, T. Force, Y. Wang, Mitogen-activated protein kinase signaling in the
heart: angels versus demons in a heart-breaking tale, Physiol. Rev. 90 (4) (2010)
15071546.
[74] N.S. Ruppender, A.R. Merkel, T.J. Martin, G.R. Mundy, J.A. Sterling, S.A. Guelcher,
Matrix rigidity induces osteolytic gene expression of metastatic breast cancer
cells, PLoS One 5 (11) (2010) e15451.
[75] A.J. Saldanha, Java Treeviewextensible visualization of microarray data, Bioin-
formatics 20 (2004) 32463248.
[76] S. Salinthone, V. Yadav, R.V. Schillace, D.N. Bourdette, D.W. Carr, Lipoic acid
attenuates inammation via cAMP and protein kinase a signaling, PLoS One 5
(9) (2010) e13058.
[77] A. Sanz-Clemente, J.A. Matta, J.T. Isaac, K.W. Roche, Casein kinase 2 regulates
the NR2 subunit composition of synaptic NMDA receptors, BMC Bioinform. 10
(Suppl. 12) (2010) S6.
[78] P. Saravanan, S.K. Venkatesan, C.G. Mohan, S. Patra, V.K. Dubey, Mitogen-
activated protein kinase 4 of Leishmania parasite as a therapeutic target, Eur. J.
Med. Chem. 45 (12) (2010) 56625670.
[79] C. Schalon, J.S. Surgand, E. Kellenberger, D. Rognan, A simple and fuzzy method
to align and compare druggable ligand-binding sites, Proteins 71 (4) (2008)
17551778.
[80] B. Schoeberl, C. Eichler-Jonsson, E.D. Gilles, G. Mller, Computational modeling of
the dynamics of the MAP kinase cascade activated by surface and internalized
EGF receptors, Nat. Biotechnol. 20 (4) (2002) 370375.
[81] E. Schwarz, F. Markus Leweke, S. Bahn, P. Li, Clinical bioinformatics for complex
disorders: a schizophrenia case study, Neuron 67 (6) (2009) 984996.
156 Q. Chen et al. / Mathematical Biosciences 262 (2015) 147156
[82] D.B. Searls, Using bioinformatics in gene and drug discovery, Drug Discov. Today
5 (4) (2000) 135143.
[83] S. Sengupta, T.R. Peterson, D.M. Sabatini, Regulation of the mTOR complex 1
pathway by nutrients, growth factors, and stress, Mol. Cell 40 (2) (2010) 310
322.
[84] L. Shao, J.J. Goronzy, C.M. Weyand, DNA-dependent protein kinase catalytic sub-
unit mediates T-cell loss in rheumatoid arthritis, EMBO Mol. Med. 2 (10) (2010)
415427.
[85] Y.B. Shan, E. Kim, M.P. Eastwood, R.O. Dror, M.A. Seeliger, D.E. Shaw, How does
a drug molecule nd its target binding site? J. Am. Chem. Soc. 133 (24) (2011)
91819183.
[86] S. Shina, L. Wolgamotta, Y.H. Yub, J. Blenisb, S.O. Yoon, Glycogen synthase kinase
(GSK)-3 promotes p70 ribosomal protein S6 kinase (p70S6K) activity and cell
proliferation, Proc. Natl. Acad. Sci. USA 108 (47) (2011) E1204E1213.
[87] G. Song, H. Zeng, J. Li, L. Xiao, Y. He, Y. Tang, Y. Li, miR-199a regulates the tumor
suppressor mitogen-activated protein kinase kinase kinase 11 in gastric cancer,
Biol. Pharm. Bull. 33 (11) (2010) 18221827.
[88] S. Sookoian, T.F. Gianotti, M. Schuman, C.J. Pirola, ENDEAVOUR software is ap-
plied to prioritize all genes of the whole genome in relation to type 2 diabetes,
Genet. Med. 11 (5) (2009) 338343.
[89] A. Subramanian, P. Tamayo, V.K. Mootha, et al. Gene set enrichment analysis: a
knowledge-based approach for interpreting genome-wide expression proles,
Proc. Natl. Acad. Sci. USA 102 (2005) 1554515550.
[90] K. Takeshita, T. Tezuka, Y. Isozaki, et al. Structural exibility regulates
phosphopeptide-binding activity of the tyrosine kinase binding domain of Cbl-c,
J. Biochem. 152 (5) (2009) 487495.
[91] C.W. Thomas, T.H. David, W.B. Ryan, S.R. Jeffrey, M.K. Robyn, A.G. Elizabeth,
H. Lan, B. Pierre, B. Lee, Computational prediction and experimental verication
of new MAP kinase docking sites and substrates including Gli transcription
factors, PLoS Comput. Biol. 6 (8) (2010) 121.
[92] T. Tian, J. Song, Mathematical modelling of the MAP kinase pathway using pro-
teomic datasets, PLoS One 7 (8) (2012) e42230.
[93] A. Torkamani, N.J. Schork, Accurate prediction of deleterious protein kinase poly-
morphisms, Bioinformatics 23 (21) (2007) 29182925.
[94] B. Trost, A. Kusalik, Computational prediction of eukaryotic phosphorylation
sites, Bioinformatics 27 (21) (2011) 29272935.
[95] P. Verdino, D.A. Witherden, W.L. Havran, I.A. Wilson, The molecular interaction
of CAR and JAML recruits the central cell signal transducer PI3K, Science 329
(5996) (2010) 12101214.
[96] M. Vieth, et al. Kinomics-structural biology and chemogenomics of ki-
nase inhibitors and targets, Biochim. Biophys. Acta 1697 (12) (2004) 243
257.
[97] Y.H. Wong, T.Y. Lee, H.K. Liang, et al. KinasePhos 2.0: a web server for iden-
tifying protein kinase-specic phosphorylation sites based on sequences and
coupling patterns, Nucleic Acids Res. 35 (Web Server issue) (2007) W588
W594.
[98] X.D. Wu, C.Q. Zhang, S.C. Zhang, Ecient mining of both positive and negative
association rules, ACM Trans. Inform. Syst. 22 (3) (2004) 381405.
[99] J.Q. Yang, H. Liu, M.T. Diaz-Meco, J. Moscat, NBR1 is a new PB1 signalling adapter
in Th2 differentiation and allergic airway inammation in vivo, EMBO J. 29 (19)
(2010) 34213433.
[100] M. Zahid, B.E. Phillips, S.M. Albers, N. Giannoukakis, S.C. Watkins, P.D. Rob-
bins, Identication of a cardiac specic protein transduction domain by in vivo
biopanning using a M13 phage peptide display library in mice, PLoS One 5 (8)
(2010) e12252.
[101] A. Zanin-Zhorov, Y. Ding, S. Kumari, M. Attur, K.L. Hippen, M. Brown, B.R. Blazar,
S.B. Abramson, J.J. Lafaille, M.L. Dustin, Protein kinase C-theta mediates neg-
ative feedback on regulatory T cell function, Science 328 (5976) (2010) 372
376.