Identifying Aging Genes in the Aging Mouse Hypothalamus Using

Gateway Node Analysis of Correlation Networks

Kathryn M. Cooper

, Stephen Bonasera

and Hesham Ali

College of Information Science and Technology, University of Nebraska at Omaha, Nebraska, U.S.A.

Department of Internal Medicine, University of Nebraska Medical Center, Nebraska, U.S.A.

Keywords: Gateway Nodes, Correlation Networks, Graph Models, Aging.

Abstract: High-throughput studies continue to produce volumes of data, providing a wealth of information that can be

used to better guide biological research. However, models that can readily identify true biological signals

from this data have not been developed at the same rate, due in part to a lack of well-developed algorithms

that can handle the magnitude, variability and veracity of the data. One promising and effective solution to

this complex issue is network modeling, due to its capabilities for representing biological elements and

relationships en masse. In this research, we use correlation networks for analysis where genes are

represented as nodes and indirect relationships (derived from expression patterns) are represented as edges.

Here, we define “gateway” nodes as elements representing genes that change in co-expression and possibly

co-regulation between states. We use the gateway node approach to identify critical genes in the aging

mouse brain and perform a cursory investigation of the robustness of these gateway nodes according to

network structure. Our results highlight the power of the gateway nodes approach and show how it can be

used to limit search space and determine candidate genes for targeted studies. The novelty of this approach

lies in application of the gateway node approach on novel mouse datasets, and the investigation into

robustness of network structures.

1 INTRODUCTION

Recently, network analysis methods have been

developed to analyze and draw signal from large,

high-throughput datasets. These methods include the

use of correlation networks, protein-protein

interaction networks, genetic interaction networks,

metabolic networks, and more. Commonly used to

describe networks of co-expression, the correlation

network model uses nodes to represent genetic

probes and edges to represent a correlated pattern of

gene expression between samples, defined by

condition, time, or other environmentally

quantifiable criterion. This technique has been

proposed for identifying differentially expressed

genes where traditional methods (such as Gene Set

Enrichment Analysis) do not always return desirable

results(Benson, Breitling 2006, Reverter, Chan

2008, Horvath, Dong 2008). As such, correlation

networks also serve as a valuable supplement to

traditional approaches.

While typically used for studying one particular

state individually, the correlation network can also

be used for comparison of states. A recent study by

Dempsey and Ali(Dempsey, Ali 2014) uses

clustering in correlation networks, particularly

clustering that identifies small, densely coA nnected

groups of genes, to compare datasets from the same

cell lines under different conditions. This analysis

revealed that clusters between states typically do not

overlap except for at a limited number of genes.

These genes that connect differentially to two

different states are termed “gateway nodes.” It has

been proposed that these gateway nodes, which are

thought to represent genes that are co-expressed with

two different sets of genes at different states, can

reveal a small, finite set of genes related to the

phenotype under scrutiny, making this approach

appealing when using high-throughput data –

typically, in studies comparing 10,000 to 40,000

genes in two or more different states, typically only

20 to 100 gateway nodes result from analysis,

depending on parameterization. Further studies on

clusters in correlation networks have found that

almost all clusters contain predicted and actual

transcription factor binding sites for common

regulatory elements(Dempsey, Ali 2014). This

M. Cooper K., Bonasera S. and Ali H..

Identifying Aging Genes in the Aging Mouse Hypothalamus Using Gateway Node Analysis of Correlation Networks.

DOI: 10.5220/0005212600360043

In Proceedings of the International Conference on Bioinformatics Models, Methods and Algorithms (BIOINFORMATICS-2015), pages 36-43

ISBN: 978-989-758-070-3

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

indicates that potentially, gene co-expression and

even possibly co-regulated could be mined from this

type of network, if such a relationship exists.

Since 1999, the network model that is

representative of biological data has found structure

and function to be related(Barabasi, Albert 1999),

particularly when the network is built using clean

data. In the protein-protein interaction networks,

high degree or hub nodes typically are more likely to

be lethal(Jeong et al. 2001, Barabasi, Oltvai 2004,

Albert 2005) clusters in these same networks

represent proteins that complex together for

functional purpose(Bader, Hogue 2003). In a genetic

interaction network, which represents the

relationships between genes when both are

simultaneously knocked out or down, the

relationship represents some measure of how

beneficial (or, more likely, detrimental) the duel

silencing is on the organism(Michaut et al. 2011).

Structures identified in these networks can lead to

identifying of genes with common pathways. The

correlation network is also known for these structure

function relationships – hubs, while not as obviously

lethal, can be enhanced to reveal lethal

properties(Dempsey et al. 2012), clusters have been

found to represent real sets of functionally related

genes(Horvath, Dong 2008), and gateway nodes

give insights into which genes play a pivotal role in

the changes in expressions from one environment to

another.

To investigate the novelty of gateway nodes in a

number of datasets with mediocre differential gene

analysis results, three datasets from varying brain

tissues of mice at 2 to 3 ages were analyzed using

the gateway nodes approach. It can be speculated

that cluster density has an impact on biological

function in correlation networks. The nature of

correlation network construction suggests that in a

network where genes are nodes and edges are

correlated patterns of expression, a clique (a network

where all nodes are connected to all other nodes) is

theoretically a more reliable or likely representation

of co-expression than a less connected cluster (also

known as a semi-clique). Consider two “clusters” of

5 genes each, one where all 5 nodes are completely

connected (10 edges) and another where the cluster

is only semi-complete (say, having 7 edges or 70%

edge density). In the example 1 below, clusters A, B,

and C all contain 7 edges – in example B, it seems

likely that edge 4—5 is incorrect, and 1-2-3-4 are

likely co-expressed. In C, it seems likely by

examining K

s that 1-2-3, 1-3-5, and 3-4-5 are all

highly correlated, but if that were truly the case, it

would stand to reason that 2-5 and 2-4 should also

be connected.

The best evidence without examining cluster

substructure is example A in this case, the densest.

To investigate the influence of density related

cutoffs on the gateway node, clusters were analyzed

using a density filter of 65-100% (65%), 75-100%

(75%) and 85-100% (85%). The goal of this study

was to analyze aging in the brain and possibly

identify the pathway players with roles in neuronal

growth and differentiation. Gateway node analysis

was again used for its application to aging and for its

design for identifying temporal expression changes.

The beauty of this and other case studies is that they

satisfy a need for application of methods to real

world data and testing of hypotheses. The results of

this study reveal a number of genes that are known

players of change in aging in the mouse brain, and

highlights how gateway nodes can be used to

identify targets of further study in similar cases.

2 METHODS

The network model used was created using data

prepared and analyzed with pairwise Pearson

Correlation (see Network Creation & Enrichment

Analysis) and was then clustered and gateway nodes

were identified (see Gateway Node Analysis).

Targets were then identified via model creation.

2.1 Network Creation and Enrichment

Analysis

Data was drawn from three microarray expression

datasets for this analysis; three were prepared in

total: (1) Cerebellum from Balb/C mice at three time

points (Young, Middle-aged, and Aged), (2)

Striatum from Balb/C mice at three time points

(Young, Middle-aged, and Aged), and (3)

Hypothalamus from male C57 mice at two time

points (Young and Middle-aged). Correlation

networks were generated using probes and

expression values from using pairwise computation

of the Pearson Correlation Coefficient (ρ) and

IdentifyingAgingGenesintheAgingMouseHypothalamusUsingGatewayNodeAnalysisofCorrelationNetworks

Figure 1: An example of how gateway nodes are scored.

In example 1, gateway nodes A and B are identified from

the overlap between network 1 (blue) and network 2 (red).

Node A has 6 edges in network 1 and 6 edges in network 2

(not including edge A-B). Node B has 3 edges in network

1 and 3 edges in network 2 (not including edge A-B).

Therefore, the total edge responsibility (total edges)

connecting gateway nodes is 18. A therefore has 12/18 or

66% gatewayness and B has 6/18 or 33% gatewayness. In

example 2, gateway node C has 4 edges in both network 1

and network 2, for a total of 8/8 or 100% gatewayness.

correlation threshold of 0.85≤ ρ ≤1.00. For each

pairwise correlation computation, hypothesis testing

using the Student’s T-test was performed; only

significant correlations (P-value <0.0005) were kept.

Gene Set Enrichment Analysis was performed using

the GeneTrail Analysis tool (http://genetrail.bioinf.

uni-sb.de/)(Backes et al. 2007). Parameters for each

analysis were set as follows:

• Organism: Mus musculus

• Analysis Type: KEGG, Gene Ontology

(manually curated only)

• P-value adjustment: FDR Adjustment

• P-value threshold: 0.05

• Minimum # categories: 2

2.2 Gateway Node Analysis

In brief, gateway nodes are identified by first

clustering networks, then networks are overlapped

and nodes that have edges in networks from both

conditions are iteratively identified. To perform

clustering, AllegroMCODE(Bader, Hogue 2003)

was used on each network under the following

parameters: Degree cutoff:10, Node Score: 0.2, K-

Core:10, MaxDepth: 10. Clustering time ranged

from 89.436 seconds (Male C57 young network) to

29,499.495 seconds (Cerebellum Balb/C young

network). Clustering correlation networks is known

to improve the lethality enrichment of high degree

nodes, largely because important hub nodes in

correlation networks are understood to be contained

within clusters. While the choice of clustering

method may vary, the lethality enrichment findings

were conducted using AllegroMCODE which

identifies many small, dense clusters. As such, this

work also includes a cursory review of how

clustering density impacts the robustness of the

gateway node. After clustering was performed,

clusters were filtered to three different arbitrarily

chosen density thresholds: clusters at or above 65%

density, at or above 75% density, and at or above

85% density. Density is defined as the number of

total edges in the network divided by the number of

possible edges – in a network with N nodes and no

duplicate edges or self-loops, the number of possible

edges is equal to [N*(N-1)]/2. As the density

threshold changes, the number of gateway nodes

present within the overlaid network changes, and as

such, it is important to consider numerous thresholds

to see if a gateway exists as an artifact of clustering

or it exists as a true gateway node, or gene that is co-

expressed with a unique group of genes in two or

more different states.

After the clustering step, networks are overlaid

on top of one another to identify gateway nodes. The

process used to identify these nodes in an automatic

way is extensively described in Dempsey and Ali

2014. Briefly, for each node in the clustered,

overlaid network, each node is first classified as

having edges in one or both networks. If the node

has edges in both networks, it is technically

considered a gateway. Scoring is then performed by

examining the number of edges per gateway per

cluster overlap versus the total number of gateway

edges (excluding intra-gateway edges). This ratio is

the gatewayness score, and reflects the

“responsibility” of each gateway in terms of how

many edges pass through that particular node from

one stage to another. An example of the difference is

shown in Figure 1. Gateway nodes were identified at

each density threshold, heretofore referenced as 65%

(at or above 65% cluster density), 75% (at or above

75% cluster density), and 85% (at or above 85%

cluster density).

After determination of gateway nodes at each

density threshold and Gene Ontology (GO)

enrichment of each gateway-connected cluster

(Ashburner et al. 2000), a model was drawn to

connect genes based on shared processes in which

BIOINFORMATICS2015-InternationalConferenceonBioinformaticsModels,MethodsandAlgorithms

Figure 2: The 65% gateway clusters from Male c57 Hypothalamus networks. Gateway nodes are in red. These are the

clusters examined using Gene Ontology in Table 2.

Table 1: Gateways by dataset. Gateways not present at 1 density only not shown. Column 1: Dataset in which gateway was

found. Column 2: Array ID for the gateway. Column 3: Gene Symbol for the gateway. Column 4: Edges running through

the gateway. Column 5: Total edges running through gateways connecting the two clusters. Column 6: Gatewayness score.

Column 7: If the gateway was found using 65% edge density clustering, the box is marked. Column 8: If the gateway was

found using 75% edge density clustering, the box is marked. Column 9: If the gateway was found using 85% edge density

clustering, the box is marked.

Dataset Array ID Gene Symbol Edges Total Edges Gateway Score 65% 75% 85%

Male c57 Hypothalamus A_51_P493919 Stk30 31 31 100.00% X X

A_51_P478132 2210019G11Rik 205 205 100.00% X X

A_52_P78684 D330040H18Rik 174 174 100.00% X X

Balb/c Cerebellum A_51_P346893 Extl1 182 182 100.00% X X

the gateway nodes are involved, if known. This

model was manually curated using the following

resources: Literature via PubMed search and review,

KEGG pathway database (Aoki, Kanehisa 2005),

and NCBI, and included regulatory relationships,

inhibitory relationships, binding relationships, etc.

This section must be in one column.

3 RESULTS

Before clustering, network sizes ranged from

38k41k nodes and 312k-7,600k edges. After

clustering, node counts ranged from 30-8k and edges

from 300-86k. Thus, network sizes changed

depending on state and tissue. As described in Table

1, the Male c57 Hypothalamus dataset contained the

fewest gateway nodes (3), with only one gene

(Stk30) found to be robust to changes at 65%, 75%,

and 85% cluster densities. The other two gateway

nodes were only found at 65% density (Tmem204,

Msx2). All three gateway nodes in this case had

scores of 100% gatewayness. The Balb/c cerebellar

dataset contained 7 gateway nodes, none of which

were robust to all three thresholds. Three were

robust to two thresholds, but only one of these are

non-RIKEN probes (Extl1). Two gateways in this set

that did not have 100% gatewayness were found

only at 65% cluster density and were shared between

IdentifyingAgingGenesintheAgingMouseHypothalamusUsingGatewayNodeAnalysisofCorrelationNetworks

Table 2: Cluster Gene Ontology Set Enrichment Analysis for Male c57 Hypothalamus dataset. Column 1: The gateway

name. Column 2: The cluster connecting that gateway – young or mid (not both combined) and edge color. Gene Ontology

enrichment (GO) or KEGG enrichment (KEGG). Column 3: GO/KEGG annotation or pathway name. Column 4:

Annotation/pathway ID. Column 5: The number of genes in that annotation/pathway name. Column 6: The p-value

associated with that enrichment. Column 7: If “down”, the cluster has fewer genes in that annotation/pathway than expected

for random. If “up”, the cluster has more genes in that annotation/pathway than expected for random. *FDR Adjustment

was used, but if a * is included in the P-value column, this indicates the annotation did not survive P-value adjustment and

the noted P-value is the unadjusted value.

Gateway

Cluster

Description

GO/

Category ID

Genes

P-value

Enrich

KEGG

Tmem204 Aged-Blue G.O. Cell GO:0005623 4 0.04* down

G.O. cell part GO:0044464 4 0.04* down

Tmem204 Yng – Green G.O. membrane GO:0016020 2 0.044* down

G.O. multicellular organismal process GO:0032501 4 0.045* down

G.O. cytoplasm GO:0005737 5 0.048* down

Msx2 Yng – Green G.O. membrane GO:0016020 2 0.044* down

G.O. multicellular organismal process GO:0032501 4 0.045* down

G.O. cytoplasm GO:0005737 5 0.048* down

Msx2 Aged-Blue KEGG ECM-receptor interaction 4512 2 0.013* down

KEGG Malaria 5144 2 0.013* down

KEGG Olfactory transduction 4740 3 0.015* down

G.O. biological regulation GO:0065007 22 0.009* down

G.O. cell GO:0005623 24 0.012* down

G.O. cell part GO:0044464 24 0.012* down

G.O. regulation of biological quality GO:0065008 4 0.012* down

G.O. multicellular organismal process GO:0032501 19 0.012* down

G.O. cellular process GO:0009987 26 0.017* Down

G.O. membrane part GO:0044425 7 0.024* Down

G.O. regulation of biological process GO:0050789 21 0.026* Down

G.O. non-membrane-bounded organelle GO:0043228 6 0.034* Down

G.O.

intracellular non-membrane-bounded

organelle

GO:0043232 6 0.034* Down

G.O. regulation of localization GO:0032879 4 0.035* Down

G.O. cellular component assembly GO:0022607 7 0.035* Down

G.O. cellular component biogenesis GO:0044085 7 0.035* Down

G.O. negative regulation of biological process GO:0048519 7 0.041* Down

G.O. membrane GO:0016020 10 0.044* Down

G.O. system process GO:0003008 12 0.045* Down

G.O. regulation of cellular process GO:0050794 16 0.046* Down

G.O. molecular_function GO:0003674 51 0.046* Up

Stk30 Aged-Blue KEGG Phagosome 4145 2 0.038* Up

G.O. binding GO:0005488 8 0.021* Up

G.O. plasma membrane GO:0005886 4 0.031* Up

G.O. membrane GO:0016020 4 0.031* Up

G.O. cytosol GO:0005829 3 0.035* Up

G.O. regulation of localization GO:0032879 2 0.038* Up

Stk30 Yng – Green G.O. organelle GO:0043226 2 0.042 Up

G.O. membrane-bounded organelle GO:0043227 2 0.042 Up

G.O. intracellular organelle GO:0043229 2 0.042 Up

G.O.

intracellular membrane-bounded

organelle

GO:0043231 2 0.042 Up

G.O. intracellular part GO:0044424 2 0.042 Up

BIOINFORMATICS2015-InternationalConferenceonBioinformaticsModels,MethodsandAlgorithms

Figure 3: The curated gateway nodes model. Genes/proteins in red are gateway nodes as listed above. Not all nodes listed

above are in the model if they do not fit or pathway information is not available. This model is not comprehensive.

two clusters – Gm8221 at 48.63% gatewayness and

Apol7c at 51.37% gatewayness. The Balb/c striatum

dataset contained the most gateway nodes at 67;

however, 19 of these were RIKEN or

unknown/unnamed genes. All the gateway nodes in

this set were not robust past 65% cluster density.

The top gateway nodes identified from Table 1

that were robust to density changes were Stk30, two

RiKEN hypothetical genes, and Extl11. Gene

Ontology enrichment analysis was performed on all

three datasets; the enrichment data was not used

particularly for gateway analysis but for consistency

and integrity of analysis to ensure biological

functions were found, indicating cluster relevance.

Results for GO enrichment on Male C57/Bl/6 mice

clusters are shown in Table 2 (results for Balb/c

datasets not shown). Based on the gateway node

analysis, Stk30 (coding for the RAGE protein) and

Extl1 are the only gateways that are non-RIKEN

genes that are robust to multi-clustering thresholds.

Based on literature collection and model curation,

Stk30 (aka RAGE) is the most upstream target that

interacts with reactive oxygen species and is also

upstream of NF-kB. Gateway node Msx2 is also

upstream of the NF-kB pathway acting as an

inhibitor of Tax gene which induces NF-kB

enhancing transcription factors. Myb is a

downstream target gateway that has ties to the

apoptotic pathway and the NF-kB pathway. Plcg1 is

acted upon by multiple proteins and goes on to

influence DAG and IP3, both (way) upstream of NF-

kB. Upstream of the same route to NF-kB as Plcg1,

the gateway node Grin2b is influenced by glucose.

4 CONCLUSIONS

Based on the gateway node analysis, Stk30 (coding

for the RAGE protein) and Extl1 are the only

gateways that are non-RIKEN genes that are also

robust to multi-clustering thresholds. The literature

collected and resulting model reveal that Stk30 aka

RAGE is the most upstream target that interacts with

reactive oxygen species and is also upstream of NF-

kB. In entirety, the model proposed above points to

activation of inflammation via NF-kB and RAGE as

a map for aging in normal Balb/C and C57 mouse

brain. A 2009 review by Kriete and Mayo confirms

a link between NF-kB activation and aging, but calls

for further investigation of the role of NF-kB outside

its well-studied role in the innate immune system

(Kriete, Mayo 2009). In our model, gateway node

Msx2 is also upstream of the NF-kB pathway acting

as an inhibitor of Tax gene which induces NF-kB

enhancing transcription factors. Myb is a

downstream target gateway that has ties to the

IdentifyingAgingGenesintheAgingMouseHypothalamusUsingGatewayNodeAnalysisofCorrelationNetworks

apoptotic pathway and the NF-kB pathway. Plcg1 is

acted upon by multiple proteins and goes on to

influence DAG and IP3, both (way) upstream of NF-

kB. Upstream of the same route to NF-kB as Plcg1

is Grin2b, influenced by glucose. All of these genes

have potential as effectors for change in the NF-kB

pathway, either upstream or downstream, but

perhaps the most important element in the model

due to gateway robustness is the RAGE protein,

encoded by gateway node Stk30. A 2003 study by

Deane et al. revealed that RAGE is a mediator of

disease-causing amyloid-beta proteins into the

central nervous system, and even suggests it as a

target for potential future therapies for Alzheimer’s

disease (Deane et al. 2003). RAGE has been found

to be up-regulated in Alzheimer’s patients (Leclerc

et al. 2009).

A 2004 study of transgenic mice with

manipulated RAGE (mAPP

/RAGE

) expression by

Arancio et al. found that pups displayed issues with

spatial memory and the NF-kB pathway is activated,

and again find it a potential target for Alzheimer’s

intervention (Arancio et al. 2004). Multiple other

evidences exist to substantiate the speculation that

RAGE plays a role in normal aging; an October

2013 PubMed search of “RAGE” + “Aging” reveals

over 100 articles relating RAGE and aging dating

back to 1999.

Application of the gateway nodes approach

allows for the utilization of the gateway nodes

approach to determine better targets for study in the

aging mouse brain. The accompanying model

provides a roadmap that points us toward RAGE,

Msx2, and Plcg1 as upstream targets for

manipulation for manipulation of expression in the

mouse brain. These genes all have indirect roles in

the NF-kB pathway; it has recently been shown that

inhibition of NF-kB in the mouse hypothalamus

resulted in a 20% increase in lifespan, improved

cognition, and levels of muscle, bone, and skin

tissue typically observed in younger mice. This

suggests that the gateway nodes approach is able to

identify genes with major roles in aging, particularly

using a robust approach. This method is able to take

sets of 30,000+ genes or gene probes and narrow it

down to only a few targets of interest, and their

potential relationships based on network modeling

of expression correlation and integration of publicly

available databases. Particularly in areas of research

where little is understood, funding is not readily

available, or resources are tight, the gateway nodes

approach can provide a robust, reproducible, and

reliable way to identify targets of interest in further

research.

Certainly, current methods for analyzing gene

expression capture just a snapshot of cellular activity

at a given time, not a dynamic process. However, the

minimal overlap of co-expression relationships in

the network form confirm that the cellular

environment is dynamic and spontaneous. This begs

the question – does a snapshot of the cell, even in

multiple replicates – accurately capture the goings-

on of cellular activity? If we were able to understand

how we got from point A to point B, we would

better understand how these gateway nodes came

about. Surely on a short-term basis gateway nodes

could arise form differential regulation of

expression, but in the long term, the question is

whether the clusters captured are a result of a short-

term cellular change or a compensatory effect of loss

of previous gene function. To improve the

dimensionality of these analyses without vastly

increasing the data load, one might consider

modifying their gene expression research design to

include 3 or more time points and to include a high

number of replicates for each time point (ideally, 5

or more). While this is certainly not always feasible

due to cost, labor, or difficulty in sample

preparation, it could be considered to help

understanding of cellular dynamics using a network

model.

ACKNOWLEDGEMENTS

This publication was made possible by Grant

Number P20 RR16469 from the National Center for

Research Resources (NCRR), a component of the

National Institutes of Health (NIH) and it's contents

are the sole responsibility of the authors and do not

necessarily represent the official views of NCRR or

NIH.

REFERENCES

Albert, R. 2005, "Scale-free networks in cell biology",

Journal of cell science, vol. 118, no. Pt 21, pp. 4947-

4957.

Aoki, K.F. & Kanehisa, M. 2005, "Using the KEGG

database resource", Current protocols in

bioinformatics / editoral board, Andreas D.Baxevanis

[et al.], vol. Chapter 1, pp. Unit 1.12.

Arancio, O., Zhang, H.P., Chen, X., Lin, C., Trinchese, F.,

Puzzo, D., Liu, S., Hegde, A., Yan, S.F., Stern, A.,

Luddy, J.S., Lue, L.F., Walker, D.G., Roher, A.,

Buttini, M., Mucke, L., Li, W., Schmidt, A.M., Kindy,

M., Hyslop, P.A., Stern, D.M. & Du Yan, S.S. 2004,

"RAGE potentiates Abeta-induced perturbation of

BIOINFORMATICS2015-InternationalConferenceonBioinformaticsModels,MethodsandAlgorithms

neuronal function in transgenic mice", The EMBO

journal, vol. 23, no. 20, pp. 4096-4105.

Ashburner, M., Ball, C.A., Blake, J.A., Botstein, D.,

Butler, H., Cherry, J.M., Davis, A.P., Dolinski, K.,

Dwight, S.S., Eppig, J.T., Harris, M.A., Hill, D.P.,

Issel-Tarver, L., Kasarskis, A., Lewis, S., Matese, J.C.,

Richardson, J.E., Ringwald, M., Rubin, G.M. &

Sherlock, G. 2000, "Gene ontology: tool for the

unification of biology. The Gene Ontology

Consortium", Nature genetics, vol. 25, no. 1, pp. 25-

29.

Backes, C., Keller, A., Kuentzer, J., Kneissl, B.,

Comtesse, N., Elnakady, Y.A., Muller, R., Meese, E.

& Lenhof, H.P. 2007, "GeneTrail--advanced gene set

enrichment analysis", Nucleic acids research, vol. 35,

no. Web Server issue, pp. W186-92.

Bader, G.D. & Hogue, C.W. 2003, "An automated method

for finding molecular complexes in large protein

interaction networks", BMC bioinformatics, vol. 4, pp.

Barabasi, A.L. & Albert, R. 1999, "Emergence of scaling

in random networks", Science (New York, N.Y.), vol.

286, no. 5439, pp. 509-512.

Barabasi, A.L. & Oltvai, Z.N. 2004, "Network biology:

understanding the cell's functional organization",

Nature reviews.Genetics, vol. 5, no. 2, pp. 101-113.

Benson, M. & Breitling, R. 2006, "Network theory to

understand microarray studies of complex diseases",

Current Molecular Medicine, vol. 6, no. 6, pp. 695-

701.

Deane, R., Du Yan, S., Submamaryan, R.K., LaRue, B.,

Jovanovic, S., Hogg, E., Welch, D., Manness, L., Lin,

C., Yu, J., Zhu, H., Ghiso, J., Frangione, B., Stern, A.,

Schmidt, A.M., Armstrong, D.L., Arnold, B.,

Liliensiek, B., Nawroth, P., Hofman, F., Kindy, M.,

Stern, D. & Zlokovic, B. 2003, "RAGE mediates

amyloid-beta peptide transport across the blood-brain

barrier and accumulation in brain", Nature medicine,

vol. 9, no. 7, pp. 907-913.

Dempsey, K., Thapa, I., Bastola, D. & Ali, H. 2012,

"Functional identification in correlation networks

using gene ontology edge annotation", International

journal of computational biology and drug design, vol.

5, no. 3-4, pp. 222-244.

Dempsey, K.M. & Ali, H.H. 2014, "Identifying aging-

related genes in mouse hippocampus using gateway

nodes", BMC systems biology, vol. 8, pp. 62-0509-8-

62.

Horvath, S. & Dong, J. 2008, "Geometric interpretation of

gene coexpression network analysis", PLoS

computational biology, vol. 4, no. 8, pp. e1000117.

Jeong, H., Mason, S.P., Barabasi, A.L. & Oltvai, Z.N.

2001, "Lethality and centrality in protein networks",

Nature, vol. 411, no. 6833, pp. 41-42.

Kriete, A. & Mayo, K.L. 2009, "Atypical pathways of NF-

kappaB activation and aging", Experimental

gerontology, vol. 44, no. 4, pp. 250-255.

Leclerc, E., Sturchler, E., Vetter, S.W. & Heizmann, C.W.

2009, "Crosstalk between calcium, amyloid beta and

the receptor for advanced glycation endproducts in

Alzheimer's disease", Reviews in the neurosciences,

vol. 20, no. 2, pp. 95-110.

Michaut, M., Baryshnikova, A., Costanzo, M., Myers,

C.L., Andrews, B.J., Boone, C. & Bader, G.D. 2011,

"Protein complexes are central in the yeast genetic

landscape", PLoS computational biology, vol. 7, no. 2,

pp. e1001092.

Reverter, A. & Chan, E.K. 2008, "Combining partial

correlation and an information theory approach to the

reversed engineering of gene co-expression networks",

Bioinformatics (Oxford, England), vol. 24, no. 21, pp.

2491-2497.

IdentifyingAgingGenesintheAgingMouseHypothalamusUsingGatewayNodeAnalysisofCorrelationNetworks