Multiple-Scale Visualization and Modeling of Biological Networks/Pathways

1
Multiple-Scale Visualization and Modeling of
Biological Networks/Pathways

• Zhenjun Hu
• Bioinformatics Program,
• Boston University, Boston, MA02215
• http//visant.bu.edu

2
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

3
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.
Hartwell LH, Hopfield JJ, Leibler S et al. From
molecular to modular cell biology, Nature
1999402C47-52
4
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-loops or multiple
edges between pairs of nodes.
Multigraph Allows multiple edges between pairs
of nodes.
Compound graph Integrates both adjacency
relations (correlations 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
5
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.
6
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?

7
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.headlesstrue -jar
  VisAnt.jar -b res/batch_cmd.txt
• Sample input/output

8
Handling large-scale networks
A functional linkage network with 15,447 nodes
and 1,722,708 edges and laid out using
elegant--gtspring-embedded relaxing, as shown at
right. The data of the network is downloaded
from http//www.functionalnet.org/mousenet/ 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
9
Handling large-scale networks
81,287
10
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 (Y2Hblue and Phylo green)
11
Advanced graphs multiscale visualization
modeling
How geographical map zooms
Countries
States
MA

TX
Cities
Blocks
12
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

13
Advanced graphs multiscale visualization
modeling

• Behind the scenes compound graph inclusive tree
  adjacency graph

A
H
B
G
C
M
inclusive tree
K
D
E
F
adjacency graph
Sugiyama, K. Misue, K. Visualization of
structure information Automatic drawing of
compound digraphs. IEEE Trans. Systems, Man, and
Cybernetics 21, 876-892 (1991).
14
Advanced graphs multiscale visualization
modeling

• Compound graph continued.

A
H
B
G
C
M
K
D
E
F

• Two restrictions
• No intersection between groups
• An rooted inclusive tree

Sugiyama, K. Misue, K. Visualization of
structure information Automatic drawing of
compound digraphs. IEEE Trans. Systems, Man, and
Cybernetics 21, 876-892 (1991).
15
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

16
Advanced graphs multiscale visualization
modeling
And life complexity seems hierarchical
Oltvai, Z.N. Barabasi, A.L. Systems biology.
Lifes complexity pyramid. Science 298,763764
(2002).
17
Advanced graphs multiscale visualization
modeling
And metabolic network seems to have a
hierarchical organization
Ravasz, E., Somera, A.L., Mongru, D.A., Oltvai,
Z.N. Barabasi, A.L. Hierarchical organization
of modularity in metabolic networks. Science 297,
15511555 (2002).
18
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
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
19
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
20
Advanced graphs multiscale visualization
modeling

• Metagraph definition

Hu Z, Mellor J, Wu J et al. Towards zoomable
multidimensional maps of the cell, Nat Biotechnol
200725547-554
21
Advanced graphs multiscale visualization
modeling

• Metanode definition

Expanded Collapsed
A
B
C
Hu Z, Mellor J, Wu J et al. Towards zoomable
multidimensional maps of the cell, Nat Biotechnol
200725547-554
22
Advanced graphs multiscale visualization
modeling

• Metaedge definition transient

Hu Z, Mellor J, Wu J et al. Towards zoomable
multidimensional maps of the cell, Nat Biotechnol
200725547-554
23
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. The
idea that a node, such as C, is known to
participate in more than one function at a given
level, is represented by displaying it in more
than one metanode. Three meta-nodes are in
expanded state and their internal network
structure is visible. (II) Meta-node G2 is
collapsed and three meta-edges H_G2 (H_B), E_G2
(E_B) and C_G2 are created based on the original
network connectivity. Meta-edge C_G2 is a
special edge because it represents the shared
components and rendered using a dashed line.
(III) Both G1 and G2 are collapsed, three
meta-edges are created, with G1_G2E_G2 H_G2,
G1_G3A_G and G3_G3C_G2. It has also been shown
here that meta-node can be embedded, with G1 and
G3 embedded in a new meta-node G4. (IV) meta-node
G4 collapsed, with a new meta-edge
G4_G2G1_G2G3_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.
24
Advanced graphs multiscale visualization
modeling
An example to use metagraph to improve the
readability and performance
Total 5,321 nodes and 33,992 edges
25
Advanced graphs multiscale visualization
modeling
An example to use metagraph to improve the
readability and performance (continued)
Total 5,321 nodes and 33,992 edges
26
Advanced graphs multiscale visualization
modeling
An example to use metagraph to improve the
readability and performance (continued)
Total 5,321 nodes and 33,992 edges
27
Metagraph for pathway visualization

• Metagraph application in pathway visualization

C
KEGG Pathway Diagram (part of G1 phase of cell
cycle)
A
B
Complex Hierarchy
E
28
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
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.
I
II
Hu Z, Snitkin ES, DeLisi C. VisANT an
integrative framework for networks in systems
biology, Brief Bioinform 20089317-325
29
Metagraph for pathway visualization

• Condition dependency

Hu Z, Snitkin ES, DeLisi C. VisANT an
integrative framework for networks in systems
biology, Brief Bioinform 20089317-325
30
Hierarchical visualization, integration modeling

• Metagraph application visualization of the
  network hierarchy

Level 3
Level 4
Module of level 3
Level 2
Level 1
Protein of level 4
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.
Hu Z, Mellor J, Wu J et al. Towards zoomable
multidimensional maps of the cell, Nat Biotechnol
200725547-554
31
Hierarchical visualization, integration modeling

• Metagraph application integrating interaction
  network with GO hierarchical modules

A
B
sequence-specific DNA binding 0(34) genes
centromeric DNA binding 6 genes
AT DNA Binding 3 genes
DNA replication origin binding 10 genes
rDNA Binding 6 genes
telomeric DNA Binding 9 genes
D
C
Hu Z, Mellor J, Wu J et al. Towards zoomable
multidimensional maps of the cell, Nat Biotechnol
200725547-554
32
Hierarchical visualization, integration modeling

• Metagraph application network of protein
  complexes

Gavin, A.C. et al. Functional organization of the
yeast proteome by systematic analysis of protein
complexes. Nature 415, 141147 (2002).
33
Hierarchical visualization, integration modeling

• Metagraph application network of protein
  complexes integrated with Y2H interactions

34
Hierarchical visualization, integration modeling

• bottom-up modeling cancer network

Goh KI, Cusick ME, Valle D et al. The human
disease network, Proc Natl Acad Sci U S A
20071048685-8690.
35
Hierarchical visualization, integration modeling

• top-down modeling disease network?cancer gene
  network

Goh KI, Cusick ME, Valle D et al. The human
disease network, Proc Natl Acad Sci U S A
20071048685-8690.
36
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 systems 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

37
Metagraph potential application in social network

• Science of Science and Innovation Policy (SciSIP)

38
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

39
Metagraph potential application in social network

• Lets model each paper (blue) as a metanode with
  authors (red) as its components and then we get a
  network of publications

Expression Analysis
Sequence Alignment
A collaboration network between different
research fields
Pathway Analysis
Biomarker Detection
40
Metagraph potential application in social network

• Lets 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
41
Acknowledge
VisANT Community


Team of Development Zhenjun Hu, Boston Univ. Evan Snitkin, Boston Univ. Yan Wang, Boston Univ. Bolan Linghu, Boston Univ. Jui-Hung Hung, Boston Univ.   Joint Developers Takuji Yamada, Kyoto Univ. Shuichi Kawashima, University of Tokyo David M. Ng, UCSC Chunnuan Chen, UCSC Changyu Fan, CCSB, Harvard Medical School   VeteransJoe Mellor,  Harvard Medical SchoolJie Wu, Boston Univ. Collaborators IBM Watson Research Laboratory KEGG Database Stuart Lab Center of Cancer System Biology Advisory Board Aravind Iyer, Computational Biology Branch, NCBI, NLM, NIH Bart Weimer, Director, Center for Integrated BioSystems, Utah State University Chris Sander, Sloan Kettering Memorial Cancer Center Daniel Segrè, Bioinformatics Program, Boston University Frederick Roth, Department of Biological Chemistry and Molecular Pharmacology, Harvard Medical School Joseph Lehár, Combinatorix, Inc Josh Stuart, Biomolecular Engineering, UCSC
Part of the support funding come from NIH Pfizer
42
Have fun with your own networks!