Gene Ontology

1
Gene Ontology
2
GO Terms Data

• Full tree available from www.geneontology.org in
  text format
• OBO file format is current, contains all terms in
  1 file
• GO format deprecated, 3 separate files
• Gene to GO category mapping available in
  Affymetrix annotations
• 6118 // electron transport // inferred from
  electronic annotation /// 6810 // transport //
  inferred from electronic annotation

3
WorkBench Model

• Select set of genes (by hand, or by process
  i.e. T-Test)
• GO panel automatically identify chip type by
  specific marker presence (or override by drop
  down)
• Check for existence of serialized tree model, if
  doesnt exist then build (from deprecated GO
  format) and serialize
• Map selected genes GO category membership onto
  GO tree
• Calculate PValue of selected genes GO category
  membership
• Compare selected genes expression levels in a
  given category to the reference genes expression
  levels this determines whether a gene is
  considered enriched or not

4
Issues - Structural

• Use of Swing DefaultTreeModel as data structure
  to represent tree limits use by other
  components (1624)
• GO tree not accessible to other components
  contained only within the GO Panel

5
Issues - Programmatic

• Of the 2.8 seconds it takes to build an in memory
  representation of the GO tree, 2 seconds is
  serializing the tree. Takes 3.9 seconds to
  deserialize tree (390ms to parse original file)
• ComputeCumulative() is recursive can overflow
  stack for deep trees (1502)
• Linear search caused by Vector.contains() at
  GoTerm190 approached 25 of runtime for large
  gene lists. Replace with something like TreeSet.
• PValue trends rendering causes OutOfMemory for
  larger sets of genes, crashing interface.

6
Issues - GUI
7
Issues GUI

• Confusing layout of components, i.e. when
  checkbox beside Reference List is checked, that
  means ignore reference list
• List of genes for a given category has no
  functionality (should be able to add to
  selections from that list too at least)
• No way to see position in GO tree for a gene in
  the table view
• Multiple progress bars, sometimes with big pauses
  in between confuse user
• Is the PValue trends graph useful?

8
Proposed fixes

• Switch to using OBO gene tree file format
• Define an interface for GO terms usage across the
  entire application and back that with standard
  data structures (not Swing support classes)
• Move GO terms into the global annotations service
• Do not serialize the GO terms tree, parse file
  every time faster, less code, less chance of
  error (i.e. if the OBO source file was updated)
• Change GUI to split pane with table view on the
  left and a dynamic tree view on the right which
  changes to display the tree for the selected gene
• Make it clear when youre using all the genes
  from the specified annotation as the reference
  list, or when youve specified a list of genes to
  override that list
• Unless P-Value trends graph provides important
  functionality, remove it or replace it by
  exposing the results of the analysis and
  displaying in a more general graphing component