Analysis of functional and pathway association of differential co-expressed genes: A
case study in drug addiction
Zi-hui Li1, Yu-feng Liu1, Ke-ning Li, Hui-zi DuanMu, Zhi-qiang Chang, Zhen-qi Li, Shan-zhen Zhang,
College of Bioinformatics Science and Technology, Harbin Medical University, Harbin 150081, China
a r t i c l e i n f o
Received 4 January 2011
Accepted 24 August 2011
Available online 28 August 2011
Gene co-expression meta-analysis
a b s t r a c t
Drug addiction has been considered as a kind of chronic relapsing brain disease influenced by both
genetic and environmental factors. At present, many causative genes and pathways related to diverse
kinds of drug addiction have been discovered, while less attention has been paid to common mechanisms
shared by different drugs underlying addiction. By applying a co-expression meta-analysis method to
mRNA expression profiles of alcohol, cocaine, heroin addicted and normal samples, we identified signif-
icant gene co-expression pairs. As co-expression networks of drug group and control group constructed,
associated function term pairs and pathway pairs reflected by co-expression pattern changes were dis-
covered by integrating functional and pathway information respectively. The results indicated that respi-
ratory electron transport chain, synaptic transmission, mitochondrial electron transport, signal
transduction, locomotory behavior, response to amphetamine, negative regulation of cell migration, glu-
cose regulation of insulin secretion, signaling by NGF, diabetes pathways, integration of energy metabo-
lism, dopamine receptors may play an important role in drug addiction. In addition, the results can
provide theory support for studies of addiction mechanisms.
? 2011 Elsevier Inc. All rights reserved.
Drug addiction has been considered as a kind of chronic relaps-
ing brain disease characterized by drug craving, drug abuse and
abstinence syndrome, which has been a worldwide issue . It is
influenced by both genetic and environmental factors, in which ge-
netic factor occupies 40–60% . Prepublished work always fo-
cused on single-type drug addiction such as morphine or
heroine; however the common mechanisms shared by diverse
drugs underlying addiction have been paid little attention to. For
each kind of addictive drug which has its distinctive pharmacody-
namics effect could cause some similar abstinence syndromes,
such as negative emotion, extension of sensitization, and associa-
tive learning process induced by drug condition, studies of com-
mon mechanisms of drug addiction will contribute to the control
and cure of addictive diseases. In 2005, Nestler elaborated the com-
mon mechanisms of four drugs, namely nicotine, alcohol, opium
and hemp combining with existing studies . In 2008, Li et al.
identified 396 genes related to at least two kinds of drug addictions
as well as five signaling pathways involved in the common mech-
anisms of four drug addictions . The high throughput gene
microarray technology is an efficient method used to analyze gene
expression level under different conditions. In previous work, the
analysis of gene expression profiles had been widely used to study
cancer and other complex diseases, and their approaches mainly
focused on the identification of differential expression genes. How-
ever, differential expression analysis which could only distinguish
single gene related to diseases can hardly reflect cooperative rela-
tions among genes. In order to settle this problem, constructing
gene co-expression network based on co-expression analysis could
identify functionally related genes, i.e. they may functioning to-
gether, and take part in the processes of disease occurrence and
Since genes involved in addiction mechanism are in extensive
wide range, single data and research method may result in a bias.
In order to overcome the limits of single data, it is better to com-
bine several independent datasets for analyzing common mecha-
nisms of drug addiction. Meta-analysis is the method that
analyzes and summarizes multiple collected data by statistical
methods aiming to provide quantified averaging effects to a ques-
tion; whose advantage is that strengthen the credibility of conclu-
sions by comprising more samples and prevent the inconsistency
of results. It takes results from several independent studies on
same project and uses appropriate statistical methods for system-
atical, objective and quantitative analysis on them. Since there is a
large range of genes involved in addiction mechanisms, single
1532-0464/$ - see front matter ? 2011 Elsevier Inc. All rights reserved.
E-mail address: firstname.lastname@example.org (Y. Xu).
Journal of Biomedical Informatics 45 (2012) 30–36
Contents lists available at SciVerse ScienceDirect
Journal of Biomedical Informatics
journal homepage: www.elsevier.com/locate/yjbin
experiment datasets and research methods may result in a bias. In
order to overcome the limits of single experiment datasets, it is
better to combine several independent datasets for analyzing
common mechanisms of drug addiction. Choi et al.  applied a
meta-analysis method on analyzing expression profiles and de-
tected differential expression genes together with correlation coef-
ficient between genes based on cancer microarray expression data
in 2003, which demonstrated meta-analysis as a efficient analysis
method for multiple datasets.
In this study, we split each dataset into a normal and a tumor
sub datasets, and constructed co-expression network respectively.
In each network, co-expressed gene pairs were identified, which
were mapped to GO terms and biology pathway. Specifically, we
selected functional differential associated pairs and pathway rele-
vant pairs which changed greatly under drug addiction and normal
status, and constructed functional association network and path-
way association network. Afterward, we analyzed their roles in
drug addiction process as well. The results indicated that respira-
tory electron transport chain, synaptic transmission, mitochondrial
electron transport, signal transduction, locomotory behavior, re-
sponse to amphetamine, negative regulation of cell migration, glu-
cose regulation of insulin secretion, signaling by NGF, diabetes
pathways, integration of energy metabolism, dopamine receptors
may play an important role in common mechanisms of three kinds
of drug addictions. In addition, the results can provide theory sup-
port for studies of addiction mechanisms.
2. Data and methods
2.1. Data collection, preprocessing and cross-platform gene mapping
Gene expression datasets were downloaded from Gene Expres-
sion Omnibus, including GDS2841 (GSE4494, GPL85), GDS255
(GSE340, GPL85) and GSE13166 whose strains were all rattus.
GDS2841 which contains 59 microarrays was designed for the
analysis of various central nervous system (CNS) regions (frontal
cortex, caudate–putamen, hippocampus, amygdala, accumbens)
of inbred alcohol-preferring (29 microarrays) and non-alcohol pre-
ferring (30 microarrays) rattus; GDS255 comprises 10 microarrays
related to different regions in brains (Amygdala, caudate putamen,
nucleus acumbens, prefrontal cortex, and ventral tegmental area)
of cocaine-treated and saline control animals; GSE13166 contains
11 microarrays of medial prefrontal cortex gene expression, six
of which were treated with heroin and the others were saline con-
trol. All data were preprocessed as follow: mapping each clone to
Gene ID, filling the missing values, standardizing between and in
all microarrays, etc. Gene annotation information such as genes
and their corresponding Entrez IDs and Gene Ontology (GO) func-
tional terms  were obtained from gene2go downloaded from
NCBI Entrez Gene ftp. And relationships between GO terms rooted
in the file go_daily-termdb-tables were downloaded from Gene
Ontology database ftp. Both of the two files above were down-
loaded in 2010, 4. The data of pathways came from Pathway Inter-
action Database (PID) , integrating pathway information from
NCI-Nature Curated, Biocarta and Reactome databases. We used
the file in XML format, PID specific, in our study which was down-
loaded in 2009, 12. The homologous information of Homo sapiens
and rats was obtained from NCBI homologene database in 2010, 2.
Thus, we collected expression datasets of alcohol, cocaine and her-
oin addiction, as well as the function and pathway information for
the subsequent studies.
2.2. A meta-analysis of gene co-expression
We applied a meta-analysis method to integrate expression
profiles above, samples of which were various drug addiction
types. In this method, correlation between every two genes was fi-
nally represented by an effect size that was transformed from Pear-
son Correlation Coefficient (PCC); l stands for the average effect
size while the observed effect size and sampling error are repre-
sented by zkand s2
tively, for all k independent datasets. And s2represents the
variability between datasets.
kwhich is also the within-study variance, respec-
zk¼ lkþek; ek? Nð0;s2
lk¼ l þ dk; dk? Nð0;s2Þ;
The PCC between gene x (gx) and y (gy) in the k -th dataset is rk, and
then converted into zkby Fisher’s r-to-z transformation.
1 þ rkðgx;gyÞ
1 ? rkðgx;gyÞ
k¼ 1=ðnk? 3Þ while s2is estimated by Cochran’s
Q-statistic. The effect size zRand its variance wkare given as the fol-
kis given as s2
Consequently, we obtained the effect size zRof every pair of
genes (e.g. gene x and gene y) to represent the expression correla-
tion between them.
2.3. Identification of functional and pathway interaction pairs
Firstly, we mapped each gene to relevant Gene Ontology (GO)
terms and Pathway Interaction Database (PID), respectively. In this
way, the co-expression gene pairs were transformed to functional
interaction pairs and pathway interaction pairs. More specifically,
if gene GA was mapped to GO category C1 and gene GB was mapped
to C2, the link GA–GB was transformed to category pair C1–C2, and
co-expression link (CL). In a similar way, pathway interaction pairs
P1–P2 formed if a pair of co-expression genes was mapped to P1
and P2 separately. In the two co-expression networks which were
constructed by normal group and drug addiction group, the signifi-
cance of every two terms or pathways co-expression was measured
by hypergeometric probability density function,
P ¼ 1 ? FðxjM;K;NÞ ¼ 1 ?
M ? K
N ? i
where the result of function F(x|M,K,N) is the probability of draw-
ing up to x of a possible K items in N drawings without replacement
from a group of M objects. More specifically, in our study, M indi-
cates the amount of possible CL between any two GO term pairs
or pathway pairs in normal or drug network, K means the number
of existing CL in the network, N stands for number of possible CL be-
tween certain term pairs or pathway pairs and x means the existing
CL between the certain terms or pathway pairs. And a shows the
percentage of gene pairs used for constructing networks occupied
in all gene pairs which was set as 0.005 in our study.
It can be inferred that functional or pathway pairs have a strong
correlation, if they have a significant P-value. While analyzing
functional interaction pairs, we picked up the GO terms which
Z.-h. Li et al./Journal of Biomedical Informatics 45 (2012) 30–36
contain more than 20 genes and whose levels are between 4 and
10. Moreover, we selected the pathways contained more than five
and fewer than 100 genes. For pathways from different databases
may correspond to exact similar or identical biological process,
for example, one pathway is a sub pathway of another, so accord-
ing to the number of genes shared by this pathway pair and each
pathway contained, we calculated P-value using Fisher exact test
followed by Benjamini–Hochberg (BH) FDR correction for each
pathway pair. The pathway pairs with P < 0.05 was taken for signif-
icantly overlapped which were abandoned as well.
3.1. Gene co-expression networks
We constructed a normal network and a drug addiction net-
work by 3380 genes which were shared by the three expression
profiles. To keep the same size for the two networks, we sorted
the gene pairs by z score (see Section 2.2), and selected the top
0.5% of the pairs severally. 28,533 gene co-expression pairs were
obtained in each group (3380 ? 3379/2). To verify the top 0.5%
genes pairs were significantly co-expression, the expressions of
all genes were permuted among all samples 1000 times randomly,
and then got all false discovery rates (FDR) values <0.01. Subse-
quently, the gene pairs selected from above two groups were
mapped to Homo sapiens according to the list of homologous
genes. As a result, we totally obtained 23,724 and 23,957 human
gene co-expression pairs in drug and normal group, respectively,
which were used for constructing two co-expression networks
(see Table 1).
3.2. Functional association network
According to the P-values of hypergeometric distribution test,
we picked up all functional relevant pairs with P-values lower than
0.05 which indicates a significant association for functional associ-
ation network construction. There were 74 functional pairs (edges)
and 55 functional terms (nodes) met the threshold requirement.
We manually classified major functional terms into 11 function
categories sorted by their mean degree (see Table 2).
As shown in Fig. 1, there are nine terms linking with more than
five other biological functional terms, which are regarded as ‘hubs’,
including respiratory electron transport chain (GO:0022904), syn-
aptic transmission (GO:0007268), mitochondrial electron trans-
port, NADH to ubiquinone (GO:0006120), central nervous system
development (GO:0007417), signal transduction (GO:0007165),
(GO:0030900), response to amphetamine (GO:0001975), negative
regulation of cell migration (GO:0030336). These highly connected
terms may play an important role in real biological networks,
whose anomalism will influence other functions to some extent.
Biologically, three kinds of drugs abuse possibly disorders some
of these ‘hub functions’ and subsequently causes abstinence
syndrome shared by the drugs. So it is important to analyze these
function terms and their neighbor terms in order to find the com-
mon mechanisms of drug addictions.
The result suggests that these functions play a significant role in
linking different functional sub networks and have an influence on
other functions in a certain degree. Furthermore, we found
that three of the nine functions, namely, neuron migration
(GO:0001764), response to amphetamine (GO:0001975) and re-
sponse to nutrient levels (GO:0031667), not only with solid links
but also dashed links (see Fig. 1). It represented that these 3 biolog-
ical processes contributed in maintaining functions in normal
bodies as well as changing cellular state when addicted to drugs.
Take response to amphetamine (GO:0001975) for instance, it solid
linked to many function terms which involved in ion or electron
transport (GO:0006813, GO:0055085, GO:0022904, GO:0006629),
therefore we can infer they may had cooperation with the term
when getting addiction; Meanwhile, this GO term dashed linked
with the biological process, neuron migration, it may participate
in maintaining normal bodies (see Fig. 1).
Among all functional association pairs whose P-value was lower
(GO:0022904) had the most biological process terms associating
with it (14 in total). In addition, five of all its associated pairs were
solid linked hub nodes, that is to say, these functional relationships
might form or enhance significantly when getting addiction. So as
we know, the respiratory pathways of glycolysis and the mitochon-
drial electron transport chain are ubiquitous throughout nature,
which are essential for both energy provision in heterotrophic cells
and a wide range of other physiological functions . Its associating
biological processes include response to nutrient or amphetamine
(GO:0007584, GO:0010033, GO:0001975), ion transmembrane
mission (GO:0007268, GO:0007165, GO:0050885), blood coagula-
tion (GO:0007596, GO:0030168, GO:0050900), etc.
3.3. Pathway association network
We collected 1025 pathways from Pathway Interaction Data-
base (PID). As the method described above, 99 pathways pairs were
selected using 0.01 as cutoff value, which included 78 unique path-
ways. As can be seen in Fig. 2, there are two kinds of special nodes
in the network, one is the nodes with high degree, such as electron
transport chain, glucose regulation of insulin secretion, signaling
by NGF, diabetes pathways, integration of energy metabolism, neu-
rotrophic factor-mediated Trk receptor signaling, dopamine recep-
tors, which all connected with more than six other pathways; and
the other is the nodes with low degree but linked to the nodes with
high degree, such as transcription factor Creb and its extracellular
signals, IFN-gamma pathway, bioactive peptide induced signaling
pathway, role of calcineurin-dependent NFAT signaling in lympho-
cytes, regulatory of ck1/cdk5 by type 1 glutamate receptors. They
all connected with four or more other pathways of high degree.
They may be associated with addiction directly or be a bridge that
was used for the communication between pathways of high
3.4. Compare with another data integrative method
By far, there are some methods or strategies for microarray-
integrated analysis. One main category of them is to use statistical
techniques, such as the meta-analysis, which combines results
generated by several studies with a same research purpose. In con-
trast to the meta-analysis method, another main category of inte-
grative strategies is to directly integrate microarray data got from
different studies into a new expression matrix. In order to compare
the effects of different methods of integrating multiple microarray
Microarray datasets included in our study.
AuthorMicroarray Platform No. of
AlcoholStrother et al.
Bonate et al.
Vrana et al.
Z.-h. Li et al./Journal of Biomedical Informatics 45 (2012) 30–36
datasets on identifying differential co-expressing gene pairs, we
also tried median rank scores (MRS) in our analysis.
The basic idea of MRS is to replace the expression values of a
single study by median expression values of a ‘‘reference’’ study.
In the first step, one of the microarray datasets is selected as a ref-
erence set where the median-expression value of each gene are
computed and stored as a vector in ascending order. Secondly,
for each microarray of non-reference set, the relative ranks of the
expression values are calculated and an expression value with rank
n is replaced by the nth element of the sorted median values vector
Afterintegrating microarraydatasets,we calculatedPCCvalueof
each gene pair and completed the remaining steps described in the
methods section. When constructing functional association net-
work and pathway association network, we took the same cut-off
values (P < 0.01) and obtained much more functional term pairs
(6457 and 6469 pairs in normal and disease functional network,
respectively). Since such a huge amount of association pairs made
it difficult to continue our analysis, we introduce a much more rig-
orous threshold to screen the most significant functional associa-
tion pairs (P < 10–7), and still a large number of pairs were left
(209 nodes and 459 edges). In functional association network, there
are 10 terms linking with more than 10 other biological functional
terms, seven of which were also discovered by the meta-analysis
method, such as nerve growth factor receptor signaling pathway
(GO:0048011), signal transduction (GO:0007165), platelet activa-
tion (GO:0030168) and axon guidance (GO:0007411) (see Table 3).
In pathway association network, 47 pathways pairs were se-
lected by the same threshold (P < 0.01). Nine pathways were
highly linked with more than 10 other pathways, and four of
which were already discovered by the meta-analysis method
(see Table 4). Parts of the other GO terms and pathways which
did not overlap our previous results, such as TRKA receptor sig-
naling pathway, ornithine metabolism and regulation of ornithine
decarboxylase, have been found had relationships with drug
Major GO terms in function association network.
ID GO term nameDegreeMean
Respiratory electron transport chain
Mitochondrial electron transport, NADH to ubiquinone
Potassium ion transport
Nerve growth factor receptor signaling pathway
Small GTPase mediated signal transduction
G-protein signaling, coupled to cAMP nucleotide second messenger
Central nervous system development
Blood vessel development
Cerebral cortex development
Negative regulation of cell migration
Regulation of G-protein coupled receptor protein signaling pathway
Regulation of translation
Positive regulation of glucose import
Negative regulation of caspase activity
Response to amphetamine
Response to organic substance
Response to nutrient levels
Ion transmembrane transport
Response to nutrient
Response to estradiol stimulus
Response to glucocorticoid stimulus
Response to testosterone stimulus
Response to retinoic acid
Xenobiotic metabolic process
Lipid metabolic process
Learning or memory
Z.-h. Li et al./Journal of Biomedical Informatics 45 (2012) 30–36
Fig. 1. Functional association network. The yellow squares represent the function terms which associated with more than five other terms; the green circles represent
function terms which associated with less than five other terms; the pink solid-lines represent association links which meet the criteria in drug group; the pink dashed-lines
represent association links which meet the criteria in normal group. (For interpretation of the references to color in this figure legend, the reader is referred to the web version
of this article.)
Fig. 2. Pathway association network. The red squares represent the pathways with high degree, and the yellow ones represent the pathways linked with those pathways of
high degree. The purple solid lines represent association links in drug group, and the purple dash lines represent association links in normal group. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)
Z.-h. Li et al./Journal of Biomedical Informatics 45 (2012) 30–36
4. Discussion and conclusions
By applying a co-expression meta-analysis method for analyzing
mRNA expression profiles of alcohol, cocaine, heroin addicted and
normal samples, significant gene co-expression pairs were identi-
fied, including 23,724 gene pairs in drug addicted group and
23,957 ones in normal group. As co-expression networks of drug
pairs and pathway pairs reflected by co-expression pattern changes
respectively with co-expression information, including 76 function
pairs and 99 pathway pairs. The results indicated that respiratory
electron transport chain (GO:0022904), synaptic transmission
(GO:0007268), mitochondrial electron transport, NADH to ubiqui-
(GO:0007417), signal transduction (GO:0007165), locomotory
behavior (GO:0007626), response to amphetamine (GO:0001975),
negative regulation of cell migration (GO:0030336), glucose regula-
tion of insulin secretion, signaling by NGF, diabetes pathways, inte-
gration of energy metabolism, dopamine receptors may play an
important role in drug addiction.
The synaptic transmission in central nervous system can be en-
hanced by nicotine. DS McGehee and colleagues revealed that nic-
otinic acetylcholine receptors could enhance fast excitatory
transmission in central nervous system by presynaptic receptors.
In addition, nicotine from tobacco had an influence on cognition
by enhancing synaptic transmission . It have been verified that
drugs do harm to central nervous system by affecting monoamin-
ergic activity, which may play an essential role in neurotransmis-
sion. And alcohol exposure will lead to mental retardation .
Fetal cocaine exposure has also been linked to numerous
nervous system development
abnormalities in neurologic, arousal, attention, and neurophysio-
logical function . Zhou and his colleagues confirmed that alco-
hol affects innervations to the developing brain by 5-HT which is a
differentiation signal for forebrain development .
Negative regulation of cell migration is aroused when getting
drug addiction, which is consistent with the fact that cocaine re-
sults in deficits in GABAergic neuronal populations and decreases
tangential neuronal migration . That signal transduction cas-
cades are responsible for alcoholic damage has been described.
Alcohol is linked to several pathologies like neurotoxicity, alcohol
liver injury, fetal alcoholic syndrome, cardiomyopathy or cancer.
These changes linked to those pathologic processes, are related
to the alteration of intracellular signaling pathways .
In our study, a hub node, namely locomotory behavior, which is
connected with four other hubs, has not been found in drug addic-
tion before. Locomotory behavior is the specific movement in re-
sponse to external or internal stimuli that dependents upon
some combination of internal state and external conditions. As
we mentioned above, drug abuse alters cellular internal state and
disturbs various functions and pathways which are essential for
maintaining normal bodies. Therefore, we forecast that locomotory
behavior is likely affected in drug addiction for unbalanced internal
cell environment. In the mouse fibrosarcoma cell, mitochondria ac-
tively participate in tumor necrosis factor-induced necrotic cell
death which is a mitochondria-dependent cell death process .
In cancer cells, it generally exhibits an increase of glycolysis for
ATP generation due in part to hypoxia and mitochondrial respira-
tion injury . For that reason, we infer that the dysfunction of
respiratory electron transport chain and mitochondrial electron
transport may be one of the causative factors in cancer, thus, drug
abuse or chronic use, such as cocaine, alcohol and nicotine, may in-
crease the risk of canceration.
In the pathway association network, some nodes have been ver-
ified to be related with drug addiction. Electron transport chain is a
major function of mitochondria, it has been reported that drug
abuse can lead to mitochondria damage and cellular energy metab-
olism failure . Chiu et al. showed that insulin receptor signaling
played an important role on synaptic plasticity, which is included
in learning and memory mechanism of addiction . Interest-
ingly, we also found diabetes pathways in the network. Chronic
ethanol intake can impair insulin signaling, and excessive alcohol
consumption increases the risk for type 2 diabetes . As for
those nodes linked with high degree nodes, many of them are also
associated with addiction. CREB was first discovered as transcrip-
tion factors which mediates effects of cAMP pathway on gene
expression . Nestler et al. showed that up regulation of cAMP
pathway is one mechanism responsible for opiate dependence
. We also find that many pathways in our network are related
to Parkinson’s disease, such as electron transport chain, dopamine
receptor. This seems unexpected but reasonable. Both addiction
and Parkinson’s disease are partly caused by the change of neuron.
We suppose that they may share some common mechanisms,
which can be studied in the future studies.
In this paper, we compared another data integrative method
(MRS)with the meta-analysis. The resultsrevealthat MRShas some
shortcomings even if it is a useful and conveniently realized ap-
proach to multiple microarray datasets integration. Firstly, much
information maybe greatly lost in the process of replacing real gene
expressionvaluesbymedian expressionvaluesof a referencestudy.
After being selected the top 0.5% of the pairs according to their PCC
values, only 1700 genes were left, which were much fewer than our
results (3380 genes in co-expression network). Such high loss ratio
of information may reduce the reliability of its results. Secondly,
the specificity of this method is not fine. We took the same signifi-
cant threshold as the meta-analysis used, but obtained a huge func-
tional association network with 6469 edges and 493 nodes, which
Highly linked GO terms in functional association network by MRS.
ID GO term nameDegree Overlap with our
Nerve growth factor receptor
Response to organic cyclic
Cellular nitrogen compound
Highly linked pathways in pathway association network by MRS.
Pathway nameDegree Overlap with our
TRKA receptor signaling pathway
Hypoxic and oxygen homeostasis
regulation of HIF-1-alpha
Regulation of ornithine
TRKA signaling from the plasma
Signaling by NGF
Z.-h. Li et al./Journal of Biomedical Informatics 45 (2012) 30–36
cannot provide precise information to experimental scientists. By
contrast, in pathway association network, the number of edges
was close to our meta-analysis. Some functional terms and
pathways were missed by MRS but were found in our study which
were already verified to be related with drug addiction, such as
respiratory electron transport chain (GO:0022904), mitochondrial
electron transport, NADH to ubiquinone(GO:0006120),central ner-
(GO:0001975), negative regulation of cell migration (GO:0030336),
signaling by NGF and integration of energy metabolism. Conse-
quently, it suggests that the meta-analysis is more suitable and
reliable in identifying differential co-expressing gene pairs from
cross-platform microarray data.
In conclusion, our method can effectively identify drug addic-
tion related biological functions and pathways, which can provide
theoretical support for future studies of drug addictive mechanism.
 Nestler EJ. Molecular basis of long-term plasticity underlying addiction. Nat
Rev Neurosci 2001;2:119–28.
 Uhl GR. Molecular genetic underpinnings of human substance abuse
vulnerability: likely contributions to understanding addiction as a mnemonic
process. Neuropharmacology 2004;47(1):140–7.
 Nestler EJ. Is there a common molecular pathway for addiction? Nat Neurosci
 Li CY, Mao X, Wei L. Genes and (common) pathways underlying drug addiction.
PLoS Comput Biol 2008;4:e2.
 Choi JK, Yu U, Kim S, Yoo OJ. Combining multiple microarray studies and
modeling interstudy variation. Bioinformatics 2003;19(Suppl 1):i84–90.
 Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, et al. Gene
ontology: tool for the unification of biology. The gene ontology consortium.
Nat Genet 2000;25:25–9.
 Schaefer CF, Anthony K, Krupa S, Buchoff J, Day M, Hannay T, et al. PID: the
pathway interaction database. Nucleic Acids Res 2009;37:D674–9.
 Fernie AR, Carrari F, Sweetlove LJ. Respiratory metabolism: glycolysis, the TCA
cycle and mitochondrialelectron
 Warnat P, Eils R, Brors B. Cross-platform analysis of cancer microarray data
 Berhow MT, Russell DS, Terwilliger RZ, Beitner-Johnson D, Self DW, Lindsay
RM, et al. Influence of neurotrophic factors on morphine- and cocaine-induced
biochemical changes in the mesolimbic dopamine system. Neuroscience
 Hannila SS, Kawaja MD. Nerve growth factor-mediated collateral sprouting of
central sensory axons into deafferentated regions of the dorsal horn is
transport.CurrOpin Plant Biol
classificationof phenotypes. BMC
enhanced in the absence of the p75 neurotrophin receptor. J Comput Neurol
 Slotkin TA, Seidler FJ, Whitmore WL. Effects of maternal methadone
administration on ornithine decarboxylase in brain and heart of the
offspring: relationshipsof enzyme
impairment in the rat. Life Sci 1980;26:861–7.
 McGehee DS, Heath MJ, Gelber S, Devay P, Role LW. Nicotine enhancement of
fast excitatory synaptic transmission in CNS by presynaptic receptors. Science
 Herlenius E, Lagercrantz H. Development of neurotransmitter systems during
critical periods. Exp Neurol 2004;190(Suppl 1):S8–S21.
 Chiriboga CA. Fetal alcohol and drug effects. Neurologist 2003;9:267–79.
 Zhou FC, Sari Y, Powrozek TA. Fetal alcohol exposure reduces serotonin
innervation and compromises development of the forebrain along the
serotonergic pathway. Alcohol Clin Exp Res 2005;29:141–9.
 Crandall JE, Hackett HE, Tobet SA, Kosofsky BE, Bhide PG. Cocaine exposure
decreases GABA neuron migration from the ganglionic eminence to the
cerebral cortex in embryonic mice. Cereb Cortex 2004;14:665–75.
 Zima T, Kalousova M. Oxidative stress and signal transduction pathways in
alcoholic liver disease. Alcohol Clin Exp Res 2005;29:110S–5S.
 Goossens V, Stange G, Moens K, Pipeleers D, Grooten J. Regulation of tumor
necrosis factor-induced, mitochondria-
dependent cell death by the electron flux through the electron transport
chain complex I. Antioxid Redox Signal 1999;1:285–95.
 Xu RH, Pelicano H, Zhou Y, Carew JS, Feng L, Bhalla KN, et al. Inhibition of
glycolysis in cancer cells: a novel strategy to overcome drug resistance
associated with mitochondrial respiratory defect and hypoxia. Cancer Res
 Virmani A, Gaetani F, Imam S, Binienda Z, Ali S. The protective role of L-
carnitine against neurotoxicity evoked by drug of abuse, methamphetamine,
could berelatedto mitochondrial
 Chiu SL, Chen CM, Cline HT. Insulin receptor signaling regulates synapse
 He L, Simmen FA, Mehendale HM, Ronis MJ, Badger TM. Chronic ethanol intake
impairs insulin signaling in rats by disrupting Akt association with the cell
membrane. Role of TRB3 in inhibition of Akt/protein kinase B activation. J Biol
 Mayr BM, Canettieri G, Montminy MR. Distinct effects of cAMP and mitogenic
signals on CREB-binding protein recruitment impart specificity to target gene
activation via CREB. Proc Natl Acad Sci USA 2001;98:10936–41.
 Nestler EJ, Alreja M, Aghajanian GK. Molecular control of locus coeruleus
neurotransmission. Biol Psychiatry 1999;46:1131–9.
 Kimpel MW, Strother WN, McClintick JN, Carr LG, Liang T, Edenberg HJ, et al.
Functional gene expression differences between inbred alcohol-preferring and
-non-preferring rats in five brain regions. Alcohol 2007;41:95–132.
 Bonate PL, Swann A, Silverman PB. Behavioral sensitization to cocaine in the
absenceof altered braincocaine
 Kuntz-Melcavage KL, Brucklacher RM, Grigson PS, Freeman WM, Vrana KE.
Gene expression changes following extinction testing in a heroin behavioral
incubation model. BMC Neurosci 2009;10:95.
activityto doseand to growth
circuit function invivo. Neuron
Z.-h. Li et al./Journal of Biomedical Informatics 45 (2012) 30–36