1. Trang chủ >
  2. Công Nghệ Thông Tin >
  3. Kỹ thuật lập trình >

2 Metagraphs: Representing Networks of Networks

Bạn đang xem bản rút gọn của tài liệu. Xem và tải ngay bản đầy đủ của tài liệu tại đây (7.45 MB, 282 trang )

11 Analysis Strategy of Protein–Protein Interaction Networks


Fig. 14. Illustration of the multi-scale visualization using metagraph. Note that node E has two instances in the inclusive

tree. (a) A network where an edge represents the inclusive relationship such as F belongs to M2, E is part of M2 and M3. (b)

A network with adjacency relations. (c) Integration of inclusive relations (dashed lines) and adjacency relations (solid lines).

(d) The intgrated network using metagraph (also referred as meta-network) where node E belongs to both metanode M2

and M3. (e) The same meta-network with three metanode (M1–3) collapsed, the dashed line between M2 and M3

indicates there is a shared node between two metanodes.

Fig. 15. Different work flows focused in this chapter. The solid lines represent the work flow of functional profiling where

GO annotations are used to interpret the roles of a given gene set. The dashed lines represent the work flow of the gene

set/network module enrichment analysis, where GO terms and associated genes may be used to construct the functional



Z. Hu

whose functions are already know and the task is to determine

whether they are enriched in the expression pattern. Finally, we

illustrate the automatic creation of the cancer gene network based

on the cancer network shown in Fig. 2 using the built-in VisANT

function “Create the co-metanode network.”

3.3. Network-Based

Functional Profiling

Functional profiling (63), or GO term enrichment analysis

(GOTEA), aims to determine whether particular GO terms inform

the difference of molecular phenotypes in any set of user-specified

genes, typically the coexpression modules (Fig. 15, solid lines). In a

network context, the goal is to identify biological functions for a

given subnetwork, or for a network module. Although many algorithms and tools (49, 64–75) have been developed for GOTEA, they

generally omit correlations based on disparate and varied datasets,

such as yeast two-hybrid, genetic interaction, mass spectrometry

(MS) and so on. Such relations may help to overcome some drawbacks in the current enrichment analysis. For example, one drawback

is that all terms are weighted equally (76), while in a network

module, terms annotated for highly connected genes will have

more weight than those annotated for the loosely connected genes.

Accuracy may also be improved if network type is considered; e.g.,

for a regulatory network, we probably can exclude those annotations

of metabolic processes. From this perspective, flexible annotation

schema will be needed to enable users to select subsets of GO

annotations as discussed in the section. Such flexibility could help

determine the functions of genes in a specified network.

3.3.1. Construct a Network

of Modules

Assume we have three coexpression clusters named CLUSTER_A,

CLUSTER_B, and CLUSTER_C, each contains a number of genes

as listed below, copy and paste (either through the pop-out menu, or

the key combination CTRL-C and CTRL-V) the following text into

the Add text box of VisANT’s toolbox and click the Add button (leftbottom corner indicated by the mouse cursor in Fig. 5), three metanodes will be created (Fig. 5). The Add textbox can be used to add

any type of the data whose format is supported by VisANT (Table 1).

#group Cluster_A





#group Cluster_B



#group Cluster_C


ERVWE1 RPP38 POP1 LOC100128314 TSEN34


11 Analysis Strategy of Protein–Protein Interaction Networks


Fig. 16. Network of three clusters created using Edge-List format and laid out using Circle Layout. Functions of the cluster

is predicted using hypergeometric testing.

Above text uses VisANT’s extended Edge-List format1 to create the network, which is the simplest format supported in VisANT.

It can also be used to easily add nodes (each line with the name of

one single node) or edges (each line with the name of the two nodes

separated by space or tab). Alternatively, users can load this edgelist from URL through File ! Open URL menu and enter the

URL http://visant.bu.edu/other_formats/edge_list_3_clusters.

txt (depending on the type of browsers, you may be able to paste

the above URL using the key combination CTRL-V), and follow

the instruction to achieve the same result. Once laid out using

Circle Layout, the network shall look similar as the one shown in

Fig. 16

3.3.2. Predict the Functions

of Modules Using

Hypergeometric Test

This method predicts the overall functions for a given set of genes by

checking the overrepresented GO terms associated with the genes.

Therefore the first step is to annotate the functions of the genes of

each cluster through the menu “MetaGraph ! GO Annotation

of All Nodes ! Using Most Specific GO Terms.” VisANT will


When you are uncertain about the format of edge-list, you can always export the network in the format of edgelist with the menu File!Export as Tab-Delimited File!All and follow the exported examples.


Z. Hu

Table 2

Cluster functions predicted with hypergeometric-based test

Molecular function

Biological process

Cluster_A Cytokine binding


Growth factor binding


Serine-type endopeptidase

activity (GO:0004252)

Positive regulation of immune

response (GO:0050778)

Innate immune response


Acute inflammatory response




Synaptic transmission (GO:0007268)

Exocytosis (GO:0006887)

Generation of a signal involved in

cell–cell signaling (GO:0003001)

Cluster_C Endonuclease activity



activity (GO:0016779)

tRNA metabolic process


Cellular component



automatically resolve the node names when annotating the nodes.

The same annotation menus are also available under the menu

“Nodes” which are only used to annotate the selected nodes. In

the case you have collapsed metanodes, such as for KEGG pathways,

always use the annotation options under “MetaGraph” menu. Once

the genes have been annotated using GO terms, we can easily

predict the functions using hypergeometric test through the menu

“MetaGraph ! Predict Functions of Metanodes Using GO !

Detect Overrepresented GO Terms Using Hypergeometric Test !

Start Hypergeometric Test over GO Database.” VisANT will perform the prediction for all non-embedded metanodes. For more

information, please reference the manual at http://visant.bu.edu/

vmanual/ver3.50.htm#hyper. The prediction results will be added

to the metanode as part of its description that are available as tooltips

when mouse-over the node (Fig. 16). Table 2 lists all predictions of

three clusters based on the reported created by VisANT: http://


Predictome database maintains a local copy of GO database and

the gene–GO associations are extracted from Entrez Gene database. Both data sets are being updated constantly therefore the

actual prediction results may be a little different from the results

shown in the link above. This also applies to the GOTEA algorithm

that will be illustrated later because the interactions are also being

updated from a list of interaction databases.

11 Analysis Strategy of Protein–Protein Interaction Networks


Fig. 17. Annotate the gene using the selected GO terms only. Four among the total thirteen terms are annotated for ACN9

because GO term hexose metabolic process (GO:0019318) are their child term, which will be very clear when the hierarchy

of GO:0019318 is shown in the GO explorer. Please reference http://visant.bu.edu/vmanual/ver3.50.htm for the information

of GO hierarchy visualization.

3.3.3. GOEA

Although it is common and fast to use hypergeometric test to

predict module’s function, the algorithm, however, does not take

into the account the interaction information for a given network

module. From this perspective, a new algorithm has been developed and implemented as a VisANT plugin to find overrepresented

GO terms in user-specified network modules (represented as metanodes in VisANT). The function is available under the “MetaGraph” menu. By default, the analysis will be performed for all

non-embedded metanodes; i.e., it is not performed for descendent

metanodes unless they are specifically selected. Similarly, overrepresented GO terms will be shown as a quick tip when the mouse is

passed over a node, and clicking on a node will display the hierarchy

of GO annotations in GO Explorer (Fig. 17). GOTEA also requires

genes in the modules to be annotated prior to the analysis.

For a given target GO term, the algorithm first computes the

density score of each node based on the path distance (number of

links) to other nodes in the same module, and the similarity

between its associated GO terms and the target term. The use of

a similarity score rather than an exact match enables the algorithm

to give the target term a high score so long as it is functionally

similar to the annotations of the genes in a module. The similarity

score between two terms is calculated by aggregating the semantic

contributions of their ancestor terms in the GO graph (77).

The enrichment of target term is determined using statistical


Z. Hu

measurement through permutation test over the subset of same

number of genes extracted from all known genes annotated by

Entrez Gene database (78) with appropriate false discovery rate

(FDR) (79) cutoff. Details of the algorithm can be found in the

Appendix. Related parameters, such as the cutoff and the iteration

number of the permutation test can be configured. By default, all

terms that have the associated genes for the current species will

need to be tested; users however, may select subset of term

branches in the GO Explorer to speed up the analysis.

The advantage of the algorithm over similar algorithms (such as

hypergeometric test) is reflected in the computation of the density

score, where the impact of one gene on another is a function of the

GO term similarity, and the number of links between the genes.

GO term similarity is calculated using a fuzzy search rather than a

conventional exact match (77). With such a density score, a gene

having many neighbors with similar GO terms will have more

significant contributions to the enrichment outcome; the algorithm

therefore leverages network topology, as well as the GO hierarchy.

In addition, metagraphs provide a flexible visual context to perform

analysis for hierarchically organized network modules. The function is designed for work flow shown as the solid red line in Fig. 15;

network modules need not be limited to expression profiling.

Permutation-based algorithms tend to be computationally

intensive and therefore time-consuming. In addition to the hypergeometric test-based algorithm, VisANT provides two options to

address this shortcoming. First, VisANT provides an option “Fast

GOTEA,” which only scans related GO terms for a given network

module (GO terms annotated for the genes in the module and

corresponding ancestor terms); and second, macro commands have

been created to allow the time-consuming GOTEA tasks be carried

out in the background with the command-line mode of VisANT.

Continue with the same example as in the previous session, and

load all interactions detected by the affinity technology (M0045) in

Predictome database (when VisANT is run as Applet, this can be

achieved through Interaction Statistics page as shown in Fig. 18).

Otherwise, they can also be loaded through metapod table in

VisANT (Fig. 1).

Once all interaction has been loaded, filter out all nodes that are

not in the three clusters will results in a network similar to the one

shown in Fig. 19. VisANT automatically adjust the global zoom

level when loading large interaction set. To resume the zoom level,

simply click first the Zoom Out button and then “Reset” button in

VisANT’s toolbox.

GOTEA can be performed through the menu “MetaGraph !

Predict Functions of Metanodes Using GO ! Network-based

GOTEA ! Fast GOTEA menu” and the iteration number is set

to 20,000 using the menu “MetaGraph ! Network-based

GOTEA ! Configure GOTEA.” The prediction results will be

11 Analysis Strategy of Protein–Protein Interaction Networks


Fig. 18. Total interactions available in Predictome database for Homo sapiens. Click on the number will load the

corresponding interactions in VisANT.

Fig. 19. Network modules for the three clusters with integrated interactions of M0045.

added to the metanode as part of its description that are available as

tooltips. Table 3 lists top three GO terms resulted from GOTEA

analysis for three clusters. The complete report can be found at:


It is obvious that GOTEA finds more enriched GO terms for

each cluster than hypergeometric test, which is mainly because

GOTEA uses a fuzzy searching algorithm to find those GO terms

that are semantically similar. As a result, GOTEA is much slower

than hypergeometric test, and takes about half hour to finish the


Z. Hu

Table 3

Cluster functions predicted by GO with integrated interaction of M0045

Molecular function

Biological process

Cellular component

Cluster_A Cytokine binding


Growth factor binding


Sugar binding



Cytokine biosynthetic process


Positive regulation of immune

response (GO:0050778)

Innate immune response



Extracellular space


Receptor complex


Secretory granule



Cluster_B Calmodulin binding


ATP binding


Calcium channel activity


Synaptic transmission (GO:0007268) Clathrin-coated vesicle


Neurotransmitter transport

Neuron projection



Generation of a signal involved in

Cytoplasmic vesicle

cell–cell signaling (GO:0003001)





Cluster_C Endonuclease activity



activity (GO:0016779)

ATP binding



tRNA metabolic process




DNA recombination (GO:0006310) Anchored to membrane


Cellular carbohydrate catabolic

Soluble fraction

process (GO:0044275)



Fig. 20. Use the red cancel button on VisANT status bar to cancel the computational heavy analysis.

analysis of three clusters. From this perspective, VisANT provides a

red cancel button at the right end of the status bar to cancel the

analysis, as shown in Fig. 20:

More information about GOTEA in VisANT can be found at


3.4. Network-Based

Expression Enrichment


Another typical application of the enrichment analysis is the study

of differential RNA expression patterns (e.g., tumor vs. normal)

determined by genome-wide association studies, to determine if

one or more specified gene sets (e.g., KEGG pathways) might

account for some of the differences (Fig. 2, dashed lines)

(80–83). Gene Set Encrichment Analysis (GSEA) (82) is probably

the most used algorithm in such analysis which does not take

account of prior network knowledge. Here we introduce the

11 Analysis Strategy of Protein–Protein Interaction Networks


Network Module Enrichment Analysis (NMEA) to test whether

the modules are enriched with transcriptional changes between the

control and the sample. NMEA is basically an extension of GSEA

but takes advantage of the extra information provided by network

connectivity. In VisANT, a network can be constructed using the

data from any combination of 70-odd methods (e.g., Y2H, ChIPChip, MS, and knockouts) for the interested gene lists. And modules can be easily constructed as metanodes through corresponding

menus, simple drage&dop operation from GO explorer, and

extended edge-list (http://visant.bu.edu/import#Edge) of user’s

own data.

Here we use the GO term to create the network modules and

then perform NMEA over them. This example can be carried out in

the following steps

1. Start VisANT as a local application (see Appendix for more

detail) and have an empty network for Homo sapiens.

2. Resume the zoom level by clicking first the “Zoom Out”

button and then “Reset” button in VisANT’s toolbox.

3. Click on the GO Explorer tab in VisANT’s control panel, enter

GO:0000077 in the search box at the bottom of GO explorer,

and click the “Search” button. Drag and drop the highlighted

term DNA damage checkpoint to the network to create the

metanode for GO:0000077 (Fig. 12).

4. Repeat step 3 for GO:0051320, GO:0007127 and

GO:0051318. All three metanodes have overlaps with the

first metanode of GO:0000077, move the overlapped genes

to the center of each metanode, and a metanetwork shall appear

similar to the one shown in Fig. 21 except there is no edge.

5. Mapping the expression profiles by opening the expression data

from the following address: http://visant.bu.edu/sample/

exp/p53_visant.dat using File ! Open URL menu.

The expression data shown in above link contains 22 microarray samples with mutations in P53 and 17 wild-type samples.

The data is downloaded from GSEA Web site (http://www.

broad.mit.edu/gsea/). Please reference http://visant.bu.edu/

vmanual/ver3.50.htm#Expression for the format of expression

data supported by VisANT.

An alternative way to load the expression data is copy/paste

expression data in the Add textbox of the toolbox.

6. Change the color mapped for the minimal and maximum

expression values to the light green and darker green, respectively, by clicking left/right side the color map shown in the

toolbar (Fig. 21). The color map will also be used to indicate

the relative contribution to the enrichment score within each



Z. Hu

Fig. 21. NMEA analysis for four GO modules in VisANT.

7. Select all nodes using Edit ! Select All Nodes menu

8. Query the interactions between selected nodes from Predictome database using Node(s) ! Query Internal Interactions

menu. The edges between the nodes appear as in Fig. 21.

In comparison to the network modules shown in Fig. 19 where

only a portion of the interactions are used to construct the

network modules, here we query all possible interactions in

the Predictome database.

9. Clear all selection with left-mouse clicking on empty space of

the network panel.

10. Start NMEA using Expression ! NMEA ! Start NMEA

Analyze menu. Once finished, p-value and FDR score will be

added to each metanode’s description (Fig. 21) and an html

report will generated similar to the one at: http://visant.bu.


From the report it is clear that only the process DNA damage

checkpoint (GO:0000077) exhibits the phenotypic difference in

the expression of genes between mutated and wild-type samples,

probably due to the fact that P53 plays a role in the process. As

mentioned in step 6, nodes with the darker color have more contribution to the enrichment score.

11 Analysis Strategy of Protein–Protein Interaction Networks


More information about NMEA in VisANT can be found at


3.5. Using Top-Down

Method to Model the

Cancer Gene Interaction


In this session we will illustrate how to use metagraph to build a

network of cancers based on the simple cancer-gene association,

and how this cancer network can be used to create the cancer gene

network. Follow instructions shows the detailed step how this

analysis can be carried out.

1. Construct the cancer network

1.1 Clear the network by clicking Clear button

1.2 Load the edge-list for the cancer network from http://

visant.bu.edu/other_formats/edge_list_cancers.txt using

the File ! Open URL menu. Once finished, click the Fit

to Page button on the toolbox.

The data shown in the above URL is extracted from

the work of Goh and coworkers (37). The disease is represented by disease ID and not very informative. From this

perspective, we use the ID-Mapping format (Table 1) to

add informative description for each cancer. The first few

lines of the file are shown below:

#!ID Mapping AddNewNode ¼ false

#VisANT_ID description

DOR2212 Rhabdomyosarcoma, alveolar, 268220 (3)







DOR2210 Rhabdoid tumors (3) [DOR2210]

DOR1804 Nasopharyngeal carcinoma, 161550 (3)


1.3 Similar to above step, load the ID-Mapping file from URL:


1.4 Collapse all metanodes using MetaGraph ! MetaNode

! Collapse All menu.

1.5 A dashed edge between two cancers will be created automatically if they share at least one gene.

1.6 Click the Zoom Out button on the toolbox 6 times and

then click the Fit to Page button to reduce the node size

and make it easier to examine the connections between


1.7 Layout the cancer network using the Layout ! SpringEmbedded Relaxing menu. Click the Stop Animation button whenever appropriate (Fig. 14). The cancer network

shall look similar to the one shown below:

Xem Thêm
Tải bản đầy đủ (.pdf) (282 trang)