Docstoc

Zu

Document Sample
Zu Powered By Docstoc
					Multiple-Scale Visualization and
     Modeling of Biological
      Networks/Pathways

            Zhenjun Hu

      Bioinformatics Program,
 Boston University, Boston, MA02215
        http://visant.bu.edu
                           Outlines
• Multiscale visualization & modeling using metagraph
   – Distinguished features of biological networks
   – Handling large-scale networks
   – Advanced graphs & multiscale visualization & modeling
       • Existing compound graph
       • Metagraph: an extension of compound graph, or an alternative
         of hypergraph that can be used for pictorial representation.
   – Metagraph for pathway visualization
   – Hierarchical visualization, integration & modeling
• Potential applications of metagraph for social
  networks



                                                                    2
                        Why networks
Circuit diagrams for biological networks ?
 The enthusiasm of the biological networks probably comes from the
 successful stories of the circuit diagrams in electronics.




             An early stored-program computer (left), built around 1950, used
             vacuum tubes in logic circuits, whereas modern computers use
             transistors and silicon wafers (right), but both are based on the
             same principles.

                                                                                                           3

                  Hartwell LH, Hopfield JJ, Leibler S et al. From molecular to modular cell biology, Nature 1999;402:C47-52
                                   Why graphs
Circuit diagrams for biological networks ?
  Tools for mining and visualizing cell systems has moved beyond static
  pictures of networks and links, most of them are based on the types of
  graphs listed below:




 Simple graph: contains no self-   Multigraph: Allows multiple     Compound graph: Integrates both
 loops or multiple edges between   edges between pairs of nodes.   adjacency relations (correlations
 pairs of nodes.                                                   between pairs of nodes) and inclusion
                                                                   relations among nodes (that is, simple
                                                                   nodes within a larger „compound‟ node
                                                                   such as the ellipse around the simple
                                                                   nodes, A and B). Compound nodes
                                                                   cannot intersect one another
 When knowledge is integrated:
 simple graph multigraph/hybrid graph compound graph                                             4
What features a biological network

 However, there are fundamental differences between biological
 networks and logic circuits:

 Scale: There are thousands of biomolecules, such as genes,
 RNAs, and proteins, each may have different states.
 Abstract: Each node represents thousands of copies of the same
 biomolecule.
 Dynamic: The biological networks are changing dynamically,
 components may appear or disappear under certain condition.
 (Modular): Biological networks may have a modular nature, and
 may organized in a hierarchical structure.




                                                                  5
       Handling large-scale networks
There are two key aspects need to be addressed when
  handling large-scale networks:
• System performance.
   –   Memory handling
   –   Right data structure
   –   Avoid nice drawing
   –   Compact size
   –   Batch mode
• Network readability.
   – Better zooming/layout?
   – Not much we can do?



                                                      6
       Handling large-scale networks
Batch mode. This mode reads instructions from a command file, and process the requests
    without any visual interface and user interactions, which enables VisANT to run in the
    background ( http://visant.bu.edu/vmanual/cmd.htm ).
•   Command to run (assume the command file is located under res directory and the name is
    “batch_cmd.txt”):

java -Xmx512M -Djava.awt.headless=true -jar VisAnt.jar -b res/batch_cmd.txt

•   Sample input/output:




                                                                                             7
             Handling large-scale networks

A functional linkage network with
15,447 nodes and 1,722,708
edges and laid out using elegant--
>spring-embedded relaxing, as
shown at right.

The data of the network is
downloaded from
http://www.functionalnet.org/mous
enet/ and directly loaded into
VisANT on a duocore computer
with 2G memory and win XP. Be
aware that we specified the
maximum memory size that are
available on the test machine in
the run.cmd: 1424M, which may
not be required by this network
and you can therefore reduce it in
case necessary. In addition,
VisANT can now directly read the
zip file therefore the downloaded
data is zipped. It takes 5+ hours
for the test case to finish


                                             8
Handling large-scale networks




                       81,287

                                9
    Handling large-scale networks

• So far we have discussed the solutions to improve
  system performance using the methods of the
  software engineering. But there seems no good
  solution to improve the network readability.

• We will discuss how to use the advanced graph to
  improve the network readability and system
  performance by integrating more biological
  information

                               An interaction network with 5489 nodes and 29,983 edges
                               (Y2H:blue and Phylo: green)
                                                                           10
 Advanced graphs & multiscale visualization
               & modeling
How geographical map zooms



                                  Countries



                 …    TX     MA      States


                                     Cities


                                    Blocks




                                              11
   Advanced graphs & multiscale
     visualization & modeling
Semantic zooming vs. geometric zooming
• Geometric (standard) zooming: The view depends on the
  physical properties of what is being viewed, objects change only
  their size.
• Semantic zooming: Different representations for different
  spatial scales. The objects being viewed can additionally
  change shape, details (not merely size of existing details) or,
  indeed, their very presence in the display, with objects
  appearing/disappearing according to the context of the map at
   hand.
• Biological network is much more complicated than
  geological maps


                                                              12
                       Advanced graphs & multiscale
                         visualization & modeling
        Behind the scenes: compound graph= inclusive tree + adjacency graph




A                 B
                                                      H
     C                                  G

                                M                K                             A                                                      H
                                                                                                 B
        D                   E                                                                                          G
                                                                                   C
                                                                                                                 M
                 F                    inclusive tree                                                                              K
                                                                                       D                     E


A                                                                                                 F
                  B                                    H
                                        G
    C
                                  M               K
         D                    E

                  F             adjacency graph
                                                                                                                             13
    Sugiyama, K. & Misue, K. Visualization of structure information: Automatic drawing of compound digraphs. IEEE Trans. Systems, Man, and
                                                           Cybernetics 21, 876-892 (1991).
                    Advanced graphs & multiscale
                      visualization & modeling
Compound graph continued.




            A                                                     H
                              B
                                                   G
                C
                                             M                K
                    D                    E

                              F



                                                                    A              B                                       H
                                                                                                            G
    Two restrictions                                                    C
                                                                                                      M
    1. No intersection between groups                                                                                 K
    2. An rooted inclusive tree                                             D                     E

                                                                                       F


                                                                                                                          14
 Sugiyama, K. & Misue, K. Visualization of structure information: Automatic drawing of compound digraphs. IEEE Trans. Systems, Man, and
                                                        Cybernetics 21, 876-892 (1991).
            Advanced graphs & multiscale
              visualization & modeling
•   Except the leaf node, each node in the inclusive tree can be thought as a group containing
    nodes of next detail level. From the point view of biological networks, such group can be a
    functional module, a protein complex etc.
•   And a biological network seems have a modular structure:




                                                                                              15
          Advanced graphs & multiscale
            visualization & modeling
And life complexity seems hierarchical




                                                                                                             16

                  Oltvai, Z.N. & Barabasi, A.L. Systems biology. Life’s complexity pyramid. Science 298,763–764 (2002).
                      Advanced graphs & multiscale
                        visualization & modeling
And metabolic network seems to have a hierarchical organization




                                                                                                                                 17
Ravasz, E., Somera, A.L., Mongru, D.A., Oltvai, Z.N. & Barabasi, A.L. Hierarchical organization of modularity in metabolic networks. Science 297,
                                                                1551–1555 (2002).
                     Advanced graphs & multiscale
                       visualization & modeling
It seems that we can use compound graph to turn a “hair ball” of interaction network
into a much readable network of functional modules:




                                                                                                                               18

     Tucker, C.L., J.F. Gera, and P. Uetz, Towards an understanding of complex protein networks. Trends Cell Biol, 2001. 11(3): p. 102-6
      Advanced graphs & multiscale
        visualization & modeling

• However, biological modules usually
  overlaps, because biomolecules usually play
  multiple roles. But compound graph does not
  support overlapping between groups
• But why the complicated circuit diagram in
  electronics does not have overlapping
  problem?  A biological network is an abstract network



                                                   19
     Advanced graphs & multiscale
       visualization & modeling
• Metagraph definition

   Gm  {V , E}
                                  V  {Vs ,Vm }
                                   E  {Es , Em }

                                                                                                      20

       Hu Z, Mellor J, Wu J et al. Towards zoomable multidimensional maps of the cell, Nat Biotechnol 2007;25:547-554
   Advanced graphs & multiscale
     visualization & modeling
• Metanode definition
                                                Expanded                        Collapsed

 vm Vm
                                            A               B

                                                C




 v V             v  vi i  0


                                                                                                       21

        Hu Z, Mellor J, Wu J et al. Towards zoomable multidimensional maps of the cell, Nat Biotechnol 2007;25:547-554
   Advanced graphs & multiscale
     visualization & modeling
• Metaedge definition: transient


      em  Em  evm ,v

      evm ,v  g (vm , v )


                                                                                                       22

        Hu Z, Mellor J, Wu J et al. Towards zoomable multidimensional maps of the cell, Nat Biotechnol 2007;25:547-554
                        Advanced graphs & multiscale
                          visualization & modeling
    • Metagraph illustration
Illustration of the dynamics of meta graph. (I) An eight gene
network grouped into three metanodes (G1, G2, G3), each
containing a set of genes that subserve some common function.                E               B
The idea that a node, such as C, is known to participate in more        I                                                   G2
                                                                                 H       C
than one function at a given level, is represented by displaying it G1    A                       G2         G4
in more than one metanode. Three meta-nodes are in
expanded state and their internal network structure is visible. (II)             F
Meta-node G2 is collapsed and three meta-edges H_G2 (=H_B),              G3 G        C
E_G2 (=E_B) and C_G2 are created based on the original                                             I IV
network connectivity. Meta-edge C_G2 is a special edge                                             II III
because it represents the shared components and rendered                     E
                                                                        I                    G2              G1             G2
using a dashed line. (III) Both G1 and G2 are collapsed, three                   H
meta-edges are created, with G1_G2=E_G2 + H_G2,                      G1   A
G1_G3=A_G and G3_G3=C_G2. It has also been shown here                                                       G4
that meta-node can be embedded, with G1 and G3 embedded in                       F
                                                                                                                  G3
a new meta-node G4. (IV) meta-node G4 collapsed, with a new              G3 G        C
meta-edge G4_G2=G1_G2+G3_G2. The procedures between I,
II, III and IV are reversible. This might be best explained in
terms of GO levels. For example G1, G2 and G3 might be GO
level 10 (pathway level) whereas G4 is GO level 9 etc.



                                                                                                                       23
   Advanced graphs & multiscale visualization
                 & modeling




                                                                                    24
An example to use metagraph to improve the readability and performance
                                                      Total: 5,321 nodes and 33,992 edges
         Advanced graphs & multiscale visualization & modeling
An example to use metagraph to improve the readability and performance (continued)




                                                                                         25
                                                        Total: 5,321 nodes and 33,992 edges
      Advanced graphs & multiscale visualization
                    & modeling
An example to use metagraph to improve the readability and performance (continued)




                                                                                       26
                                                         Total: 5,321 nodes and 33,992 edges
    Metagraph for pathway visualization
• Metagraph application in pathway visualization




                                         C

  KEGG Pathway Diagram
  (part of G1 phase of cell cycle)   A


 Complex Hierarchy                   B




                                         E
                                                   27
           Metagraph for pathway visualization
     • Metagraph application in pathway visualization (continued)



                                                                             Improved readability and performance with multi-scale I
                                                                             information integrated in pathway visualization using
                                                                             metagraph. Blue boxes represent the KEGG pathways;
                                                                             blue boxes with dark border are contracted metanodes
                                                                             representing a group of proteins; orange boxes with
                                                                             light border representing the protein complex, filled
                                                                             circles represent protein and open circles represent
                                                                             compounds. (I) Five signaling pathways of Homo
                                                                             sapiens visualized using metagraph, dashed lines
I                                                                            indicate that there are shared nodes. (II) Same number
                                                                             of pathways visualized as an interaction network. The
                                                                             size of the node is reduced to improve the readability.




II                                                                                                                     28

            Hu Z, Snitkin ES, DeLisi C. VisANT: an integrative framework for networks in systems biology, Brief Bioinform 2008;9:317-325
  Metagraph for pathway visualization
• Condition dependency




                                                                                                               29

    Hu Z, Snitkin ES, DeLisi C. VisANT: an integrative framework for networks in systems biology, Brief Bioinform 2008;9:317-325
        Hierarchical visualization, integration
                     & modeling
• Metagraph application: visualization of the network hierarchy
   Level 4                                                       Level 3




                           Module of level 3
   Protein of level 4                                  Level 1    Level 2




  Level 1: 1 module
  Level 2: 8 modules
  Level 3: 161 modules
  Level 4: 810 proteins. Only part of proteins are shown
          in the figure due to space limit.
                                                                                                                       30

                        Hu Z, Mellor J, Wu J et al. Towards zoomable multidimensional maps of the cell, Nat Biotechnol 2007;25:547-554
     Hierarchical visualization, integration
                  & modeling
•   Metagraph application: integrating interaction network with GO
    hierarchical modules

              A                  sequence-specific
                                 DNA binding
                                                                         B
                                 0(+34) genes


              centromeric rDNA    AT DNA telomeric DNA DNA replication
              DNA binding Binding Binding Binding      origin binding
              6 genes     6 genes 3 genes 9 genes      10 genes




              C                                                          D


                                                                                                            31

             Hu Z, Mellor J, Wu J et al. Towards zoomable multidimensional maps of the cell, Nat Biotechnol 2007;25:547-554
            Hierarchical visualization, integration
     •
                                  & modeling
           Metagraph application: network of protein complexes




                                                                                                                             32

Gavin, A.C. et al. Functional organization of the yeast proteome by systematic analysis of protein complexes. Nature 415, 141–147 (2002).
     Hierarchical visualization, integration
                  & modeling
•   Metagraph application: network of protein complexes integrated with
    Y2H interactions




                                                                      33
     Hierarchical visualization, integration
                  & modeling
• bottom-up modeling: cancer network




                                                                                                         34

             Goh KI, Cusick ME, Valle D et al. The human disease network, Proc Natl Acad Sci U S A 2007;104:8685-8690.
    Hierarchical visualization, integration
                 & modeling
• top-down modeling: disease networkcancer gene network




                                                                                                        35

            Goh KI, Cusick ME, Valle D et al. The human disease network, Proc Natl Acad Sci U S A 2007;104:8685-8690.
                     Quick summary


• Metagraph improves the network readability and system
  performance with integrated context information.
• Metagraph helps to represent the complication of the biological
  network, such as condition-dependency, combinatory control
  etc.
• Metagraph extends the system’s capability to integrate
  multiscale knowledge, making it much more practical to
  model/simulate the complexity of biological system: from cell to
  functional module, network motif, protein…




                                                                     36
      Metagraph: potential application in
               social network
• Science of Science and Innovation Policy (SciSIP)




                                                      37
      Metagraph: potential application in
               social network
• What can be expected from SciSIP?



   1. Predict potential research innovation

   2. Predict potential new cross-discipline research fields

   3. Predict potential collaboration between different research scientists

   4. and more ……




                                                                        38
       Metagraph: potential application in
                social network
• Let’s model each paper (blue) as a metanode with authors (red)
  as its components and then we get a network of publications:




                                     A collaboration network
                                 between different research fields




                                                                     39
       Metagraph: potential application in
                social network
• Let’s turn the publication network into co-author network:




                               More importantly, an author can also be
                               modeled as a metanode with educations,
                              hobbies etc. as the subcomponents, which
                              will enable us to draw the correlations from
                                          heterogeneous data



                                                                             40
                                      Acknowledge
VisANT Community

 Team of Development:                     Collaborators:                    Advisory Board:
 Zhenjun Hu, Boston Univ.                 IBM Watson Research Laboratory    Aravind Iyer, Computational Biology
 Evan Snitkin, Boston Univ.               KEGG Database                     Branch, NCBI, NLM, NIH
 Yan Wang, Boston Univ.                   Stuart Lab                        Bart Weimer, Director, Center for
 Bolan Linghu, Boston Univ.               Center of Cancer System Biology   Integrated BioSystems, Utah State
 Jui-Hung Hung, Boston Univ.                                                University
                                                                            Chris Sander, Sloan Kettering
 Joint Developers:                                                          Memorial Cancer Center
 Takuji Yamada, Kyoto Univ.                                                 Daniel Segrè, Bioinformatics
 Shuichi Kawashima, University of Tokyo                                     Program, Boston University
 David M. Ng, UCSC                                                          Frederick Roth, Department of
 Chunnuan Chen, UCSC                                                        Biological Chemistry and Molecular
 Changyu Fan, CCSB, Harvard Medical                                         Pharmacology, Harvard Medical
 School                                                                     School
                                                                            Joseph Lehár, Combinatorix, Inc
 Veterans:                                                                  Josh Stuart, Biomolecular
 Joe Mellor, Harvard Medical School                                         Engineering, UCSC
 Jie Wu, Boston Univ.


                                              Charles DeLisi


                       Part of the support funding come from NIH & Pfizer

                                                                                                      41
        Have fun
with your own networks!




                          42

				
DOCUMENT INFO
Shared By:
Categories:
Stats:
views:2
posted:8/27/2011
language:English
pages:42