Moore Notes 4 27 10

OTU Project Group Call

• Update from Tom: Slides
  • OTU pipeline
  • Simulation process (two types, one with reads that run off the end of the gene)
  • Results comparing read vs. full-length source analyses
    • Phylogenetic distance matrices: Mantel tests show moderate (but significant) positive associations
    • Overlap of OTU clusters: Sensitivity and specificity are much better than random clusters, best for small MOTHUR cutoffs
    • Adjusting the MOTHUR cutoff for reads to maximize TNR (specificity) subject to TPR (sensitivity) > 80%
      • Works as long as source clustering is at a cutoff < 0.1
      • Max accuracy at cutoff = 0.15 for source cutoffs < 0.05
    • Number of clusters: also suggests increasing the cutoff
  • Results comparing phylogenetic to percent identity distance
    • OTU pipeline underestimates total biodiversity (no false negatives)
  • Results comparing phylogenetic clusters (full length sequences) to taxonomy clusters (species, genus, phylum)
    • Genus cutoff of 0.05 is OK
    • Species will be undercounted at 0.03 cutoff
  • Results about what factors might be contributing to error in the reads vs. source analyses
    • Position in alignment: no
    • Conservation across 16S: no
    • Info in alignment column (number of reads aligned): no
    • Length: yes; reads < 100bp are contributing a lot, slightly longer reads might be

• Future work
  • Ideal adjustment of cutoff
  • How long do reads need to be? e.g. filter reads < 100bp, maybe longer
    • Relates to doing Illumina vs. 454, pyrotags vs. shotgun
  • Other methods:
    • James will explore clustering methods using alternative linkage methods in MOTHUR
    • Sam will explore RAxML (vs. fastTree)
  • Relative abundance?
  • Size of reference db?
  • Steve will work on downstream analyses of GOS/gut data
  • Tom will make an outline and folks will look for where they can contribute
  • Journal?
    • Nature Methods
    • PLoS Comp Bio
    • ISME
  • What papers do we need to cite? speak to? improve upon?

• Things to emphasize
  • Phylogenetic distance handles non-overlapping reads, gets around PCR bias
  • Modular pipeline that is easy to use, high through-put
  • Method for adjusting cutoffs