Bioinformatics and Computational Biology Projects

1
Bioinformatics and Computational Biology Projects

• HiPCAT Meeting
• September 22, 2005
• Richard H. Scheuermann
• Department of Pathology
• U.T. Southwestern Medical Center

2
Overview

• Projects
• Bioinformatics Resource Centers for Biodefense
  and Emerging/Re-emerging Infectious Disease (BRC)
• Bioinformatics Integration Support Contract
  (BISC)
• Data Analysis and Algorithm Development
• Microarray data analysis and gene ontology
  integration
• Microarray data analysis and transcription
  regulation integration
• Biological network analysis
• Yeast protein-protein interaction network
• B lymphocyte genetic interaction network
• Population genetic analysis
• Genome annotation
• Multi-dimensional fluorescence activated cell
  sorting analysis

3
Bioinformatics Resource Centers for Biodefense
and Emerging/Re-emerging Infectious Disease
(BRC)
4
BRC Program Overview

• NIAID (National Institute of Allergy and
  Infectious Diseases)
• Significant investment in sequencing pathogens
• Potential targets for diagnostics, therapeutics,
  and vaccines
• Desire to facilitate data exploration
• Desire to integrate data and tools from disparate
  sources
• Accordingly, NIAID awarded 8 BRC contracts
• 5-year contract starting June, 2004
• NIAID Program Officer - Dr. Valentina di
  Francesco
• The BioHealthBase BRC
• 17 million project
• Collaboration with Northrop Grumman IT

5
BioHealthBase BRC Overview

• Web-base System Accessible to Science Researchers
• Storage of sequences, annotation, reference
  databases, etc.
• Generation and storage of value added products
• Pathways, promoter regions, operons,
  multi-alignments, epitopes, etc.
• Deployment of available science community tools
• GBrowse, BioCyc Navigator, etc.
• Development of new analytical tools
• Development of new query and visualization tools
• GeneCards query tool with PlasmoDB-like
  interface, etc.
• Emphasis on host-pathogen interactions to support
  drug discovery and vaccine development

6
BioHealthBase BRC - Pathogen Rationale

• Each pathogen is not only a potential
  bioterrorism agent, but is also endemic or
  re-emerging in the United States
• dual purpose biodefense and public health
• At least one pathogen from NIAID Category A, B,
  and C
• Pathogens represent each of the three major
  classes of microorganism - bacteria, viruses, and
  parasites - and even plants
• Developing a database that supports these
  disparate organism types represents a significant
  challenge. However, the microorganism-independent
  database structure that we will develop will
  support the addition of other organisms in the
  future

Francisella tularensis Influenza virus
Mycobacterium tuberculosis Microsporidia Giardia
lamblia Ricinus communis
7
Bioinformatics Integration Support Contract (BISC)
8
BISC summary

• The Bioinformatics Integration Support Contract
  (BISC) was awarded to provide advanced computer
  support for the collection, integration and
  analysis of immunology research data from
  numerous, diverse sources, and their long-term
  storage in sustainable databases
• 28 million project
• Collaboration with Northrop Grumman IT
• The users of BISC include all NIAID/DAIT-funded
  programs, which includes
• basic scientific research
• clinical trials
• The Immunology Database and Analysis Portal
  (ImmPort) is being developed as the solution to
  the needs defined in the BISC RFP
• System deployment - ImmPort v1.0 by 26OCT2005

9
ImmPort System Components

• Data Warehouse
• Access restricted to DAIT/NIAID investigators
• Containing public reference data, primary
  experimental data, processed data
• Built upon a cross-disciplinary data model
  supported by a customized ImmPort Ontology
• Private Project Workspaces
• For pre-publication data storage and analysis
• PI-controlled access to collaborators
• Securely segregated from public warehouse
• Analytical Toolkit, Interfaces and Reports
• Database query interfaces
• Data analysis applications
• Journaling and reports
• User Support
• Data sharing compliance
• Educational resources
• Outreach

10
ImmPort - Principles

• Utilize open source software as much as possible
• Contribution of data to the ImmPort system should
  require minimal effort and should significantly
  enhance the research program of the investigator
  (incentive)
• Ensure that data is appropriately attributed
• Data integration is key
• Provide analytical tools - extant and novel
• Provide training in the use of the analytical
  tools
• Develop and utilize data standards, including
  controlled vocabularies

11
Public reference data

• Protein information
• Protein description
• Cross referencing IDs
• PubMed references
• Domain structure
• Motif location
• Molecular function
• UniProt

• Basic gene information
• Standard gene symbol/name
• Cross referencing IDs
• Gene location/chr. position
• GO terms
• Reference sequences
• Homologues
• NCBI Entrez

• Polymorphism information
• SNP location
• SNP allele frequencies
• SNP haplotype blocks
• Microsatellite/VNTR location
• Disease association
• NCBI dbSNP, dbMHC, UniSTS
• HapMap
• CEPH

Immune response gene
Gene expression information NCBI GEO
Metabolic pathways BioCyc, KEGG
Protein-protein interaction networks DIP, BIND,
MIPS, AfCS Y2H
Signal transduction pathways AfCS Molecule Page
network, Reactome
12
Data Analysis and Algorithm Development
13
Parallel MoNet

• Biological network analysis

14
Motivation

• The cell functions as a system of integrated
  components
• There is increasing evidence that the cell system
  is composed of modules
• A module in a biological system is a discrete
  unit whose function is separable from those of
  other modules
• Modules defined based on functional criteria
  reflect the critical level of biological
  organization (Hartwell, et al.)
• A modular system can reuse existing, well-tested
  modules
• The notion of regulation requires the assembly of
  individual components into modular networks
• Functional modules can then be assembled together
  into cellular networks
• Thus, identifying functional modules and their
  relationship from biological networks is
  important to the understanding of the
  organization, evolution and interaction of the
  cellular systems they represent

15
Definition of network modules
1
2
3
16
Edge betweeness

• Girvan-Newman proposed an algorithm to find
  social communities within human population
  networks
• Utilized the concept of edge betweenness as a
  unit of measure
• defined as the number of shortest paths between
  all pairs of vertices that run through it
• edges between modules tend to have higher values
• Provides a quantitative criterion to distinguish
  edges inside modules from the edges between
  modules

Betweenness 20
17
A new definition of network modules

• Definition of module degree
• Given a graph G, let U be a subgraph of G (U? G).
  The number of edges within U is defined as the
  indegree of U, ind(U). The number of edges that
  connect U to remaining part of G (G-U) is defined
  as the outdegree of U, outd(U).
• Definition of module
• A subgraph U? G is a module if ind(U)? outd(U).
• A subgraph is a complex module if it can be
  separated into at least two modules by removing
  edges inside it. Otherwise, it is a simple
  module.
• Adjacent relationship between Modules
• Given two subgraphs U, V? G, U and V are adjacent
  if U?V? and there are edges in G connecting
  vertices in U and V.

18
Algorithm

• 1. The G-N algorithm is run on the network and
  the order of edge deletion is obtained.
• a. The betweenness scores for each edge in the
  network is calculated.
• b. The edge with the highest betweenness is
  identified and removed from the network.
• c. Step 1 is repeated until no edges remain in
  the network.
• 2. An edge list is created in the reverse order
  of edge deletion in Step 1.
• 3. The agglomerative algorithm is initialized by
  setting each vertex as a singleton sub-graph with
  no edges. All singleton sub-graphs are labeled
  as mergable.
• 4. An edge is removed from the top of the edge
  list.
• 5. If the edge connects vertices in the same
  sub-graph, it is added to the sub-graph.
• 6. If the edge connects vertices in two different
  sub-graphs
• a. If both sub-graphs are mergable, the two
  sub-graphs are evaluated based on the module
  definition.
• i. The edge is retained if merging occurs
• 1) between two non-modules, or
• 2) between a non-module and a module
• ii. Otherwise, the two sub-graphs are set as
  non-mergable.
• b. If one of the sub-graphs is non-mergable, the
  other sub-graph is set as non-mergable.
• 7. Repeat steps 4 - 7 until no edges are left in
  the edge list.

19
Modular decomposition of the Yeast PIN

• Yeast protein interaction network
• The core protein interaction network of yeast
  from DIP database (version ScereCR20041003)
  (Xenarios et al. 2002)
• After removal of all self-connecting links, the
  final protein interaction network included 2609
  yeast proteins and 6355 interactions
• Consist of a large component of 2440 proteins and
  65 small components with size no more than 7
  proteins

20
Modular decomposition of Yeast PIN

• Result Summary
• 202 simple modules generated
• 137 simple modules from the large component
• The size of modules ranges from 2 to 201

21
Gene/protein annotation

• Gene Ontology (GO) Consortium database that
  describes genes based on their molecular
  function, biological process and cellular
  component.
• Gene example - Retinoblastoma
• Function - DNA binding
• Component - nucleus
• Process - transcription regulation regulation of
  cell cycle
• Description represents a leaf on a branch in a
  hierarchical tree indicating parent-child
  relationships (directed acyclic graph).
• Although we have used GO annotation as a means of
  gene classification, CLASSIFI can be used with
  any gene description scheme of interest.

22
Validation of modules

• Annotated each protein with the Gene OntologyTM
  (GO) terms from the Saccharomyces Genome Database
  (SGD) (Cherry et al. 1998 Balakrishna et al)
• Quantified the co-occurrence of GO terms using
  the hypergeometric distribution analysis
  supported by the Gene Ontology Term Finder
  (Balakrishna et al)
• The results show that each module has
  statistically significant co-occurrence of
  functional GO categories

23
Validation of modules
Modules with 100 GO frequency
24
Validation of modules

• P values of modules obtained based our definition
  plot against P values of the corresponding weak
  modules (two modules share most of their content
  genes)

25
Why Parallelize?
26
Time Complexity of MoNet

• Time complexity of G-N algorithm is O(M2N), where
  M is the number of edge and N is the number of
  vertices.
• Each step to delete high betweenness edge need
  O(MN).
• The time complexity of MoNet is (M2NM).

27
Why Parallelize?

• Modular decomposition of the yeast core protein
  interaction network from DIP
• Number of vertices 2609
• Number of edges 6355
• Time complexity 635522609 6355) 1.1 x 1011
• Run time on a server running Microsoft windows
  server 2003 Enterprise Edition with two Intel
  Xeon 3.2 GHz processors and 4G RAM 2 days

28
B lymphocyte Genetic Network

• Reverse engineering of regulatory networks in
  human B cells K. Basso, A.A. Margolin, G.
  Stolovitzky, U. Klein, R. Dalla-Favera  A.
  Califano Nature Genetics  37, 382 - 390 (2005)
• Cellular network from a set of 336 expression
  profiles representative of normal B cell subpops,
  various subtypes of B cell tumors and
  experimentally manipulated B cells
• MoNet Analysis
• Number of vertices 6111.
• Number of edges 64751.
• 6475126111 64751) 2.6 x 1013
• Run time on a server running Microsoft windows
  server 2003 Enterprise Edition with two Intel
  Xeon 3.2 GHz processors and 4G RAM 50 days

29
Where to Parallelize

• The calculation of all shortest path in the G-N
  algorithm to obtain the edge betweenness.