Moore Notes 8 3 11

Group Call - Protein Db Authors

• Sensitivity/Specificity analysis
  • Results look much better now that inappropriate sequences have been removed from analysis
  • Precision better in global (coverage threshold) vs. local analysis - probably due to shared domains
• Family Relationships
  • parent = longest consensus sequence, child = shorter one
  • Corresponds essentially to super- and sub-families
  • This is an annotation, not a change to the family definitions
  • Most have 0 or 1 relationship, but there is a long tail in the histogram out to >2000 relationships
    • Put histogram on log scale
  • Cytoscape visualization
    • Try other renderings that separate out subnetworks
    • Calculate summary statistics (betweenness, degree/degree centrality)
    • Look for subnetworks, maybe show one or two as examples
    • Do we have phylogenetic relationships represented here?
• Fragment analysis
  • Everything that was filtered from MCL analysis and singletons
  • Compare fragments to families
  • Do we need a coverage threshold on the fragment? (had one on fragment and on family originally)
    • Cuts number of hits in half
    • Goal: have we kicked out data that should be in the families?
    • Probably do want coverage threshold to ensure family membership
  • How does number of hits correlate with fragment length? e.g., short fragments might cover common domains or motifs
  • How does number of hits correlate with family coverage?
• GOS analysis
  • Used annotated peptides from CAMERA (hard to track down!)
  • Could use all predicted ORFs or assembled peptides from these ORF peptides (more manageable - 6 million proteins)
    • Should take about 1 week on cluster to search vs. all families
• In progress:
  • PFAM analysis
  • Newly sequenced genomes scan
• Next call: 2 weeks from now