Moore Notes 11 17 10 PI

Group Call

• PhylOTU manuscript is resubmitted

• Niche mapping figures
  • Range maps are really just for abundant organisms
  • Hard limit of occurring at 10 sites in the world to do niche mapping
    • What global abundance does this correspond to? can we quantify?
  • Worse for microbes, but happens with other organisms too
  • Would be interesting to run on OTUs, but there are a lot of issues with that
  • Omission rates figure
    • only pick up taxa that are "closely" related to things we've seen before (50% RDP cutoff)
    • Wouldn't be a problem if the rare taxa / endemics scale spatial with total richness (not likely to be true - isn't for macroorganisms)
    • Main issue is whether the patterns we describe may be biased by only using common taxa
    • Focus on order or family most likely
      • But what can we say about the range of a taxa at this high level of taxonomy?
      • Higher level (order) for summary/meta-maps vs. finer level for maps of individual taxa
      • Scale of the map is important too
  • There is literature on how biogeographic patterns change at different levels of taxonomic resolution (e.g., species/genus ratios)
    • Not just abundant organisms
    • Could we make quantitative comparisons with marine macro organisms?
      • Census of marine life data http://www.coml.org/ with data here: http://www.iobis.org/
      • Tittensor paper
      • Aquamaps.org
      • Qualitative comparisons might be sufficient
  • Will use an RDP bootstrap cutoff of 50%
    • 99% of sequences are v6 pyrotags, and this is a good cutoff for them
    • agrees well with Greengenes
  • Guillaume is error checking the code
  • Writing up paper

• AMPHORA-2
  • BLAST step to pull out sequences in a family
    • HMMR3 is fast enough to use the profile HMM, but is less sensitive then
    • Currently using MCL clustering to identify groups within list of sequences and use one sequence per group
    • Other ideas:
      • Build tree and pick representatives using maxPD
        • Can over sample clades with long branches
      • Emit sequences from HMM
        • Do the simulated sequences represent the family well? (since sequences are "averages", not real sequences)
      • Kimmen's software should do this, but is geared towards clustering at a higher level (subfamilies within a superfamily)
        • May have some of the same problems as maxPD
      • Prune tree from inside out - need an algorithm
      • Would node imbalance statistic be useful here?
      • Build a simple, fast feature-based classifier
    • Goal: to pick sequences that are optimal for detecting different phylogenetic lineages within a family
  • Focus on input and outputs for each module, so that other methods can be swapped in at any step
    • Build a modular workflow that can be easily extended
  • Pplacer is running to build trees
  • Next
    • More families
    • Better modules
  • Steve can be a tester when the time comes