Moore Notes 8 21 13

Group Call

• Participants: Katie, Jonathan, Guillaume, Josh, Ladan, Stephen, Sarah, Tom, Dongying

• EMP sequencing
  • Re-requested sample list from Gilbert lab (Sarah Owen will send it soon)
    • Need to select samples to be complementary to Tera Oceans and Gulf of Mexico projects
    • E.g., fill in global coverage to enable a global map or pick a different ecosystem with interesting ecology
  • They are going to start doing rapid-seq on the Hi-seq machine, which will give 250PE
  • Stephen: can we get the insert short? Jonathan: Yes with a little variation in the insert length
  • Jonathan: showed BGI the Argon price list

• SFams updating issues
  • See last week's notes for summary of the issues
  • Proposed fix:
    • Wipe round 2 results
    • Do a fast blast search before hmm search
  • How much time to invest in continuing to update the database?
    • How automated can we make this?
    • What about people who will come to depend on it?
    • Important to do it if we need it for our biogeographical studies
  • Any traction on getting some one else to take this on?
    • Jonathan: no one seems ready to take on a more rapid updating approach yet (need more money)
  • Stephen: lots of ideas for further improvements
    • Pangenome idea to ignore identical or similar genomes
    • Improving MCL step, or adding steps afterwards to filter, split, combine families
    • How to include metagenomic data in the build?
  • Tim will implement Stephen's solution to the duplicate sequence problem
    • Then reanalyze the round 2 genomes to see if this solves the problem
    • If two-step search is used in QC, then some other bad families might get better
    • Think about an automated system for handling bad families
    • Stephen: probably was also a problem between round 0 and round 1
    • Katie: Why is this happening?
      • Small families with a bogus member
      • Large families that are diverse and therefore have a low information content HMM
      • Sequence that HMM doesn't classify back is a bogus member (problem with MCL?)
      • Dongying: Connections can get added in inflation step of MCL
      • Ladan: What about using a lower inflation parameter? Tom: Trying to get larger clusters, also this didn't help much overall
    • Stephen: What fast blast tool do people recommend?
      • USearch and CDHit
      • Already looked at RAPSearch, LAST