Moore Notes 4 7 10

Group Call

• Steve will check with PLoS about our paper

• Update from Steve
  • GOS analysis
    • Guillaume and Dongying produced new marker gene families for GOS
      • Now have good archaeal markers (old ones were mostly hitting bacteria)
    • Steve will run PD analyses on GOS with these markers
    • Can now search, align, mask other data sets
  • Does phylogenetic distance in tree predict a pair of genome's similarity in terms of 16S copy number?
    • KP: take a look at Matt Hahn's CAFE program
    • Can adjust relative abundance estimates to account for copy numbers
    • JE: would be interesting to do a phyla level analysis of normalized vs. non-normalized vs. protein marker genes (single copy)
      • Did this on HOT/ALOHA and human distal gut, and saw a big difference in HOT/ALOHA (Prochlorococcus goes from rare to abundant)
      • Will do GOS, shotgun (better) and PCR
    • JE: relative abundance could also be adjusted for genome size (since genome size may affect probability of getting a read in 16S)
      • AD: could estimate genome size by proportion of reads hitting genome assemblies (compare single vs. multiple copy genes)
      • SK: recent ISME paper about correcting for genome size (on citeulike): http://www.citeulike.org/group/6072/article/6912281
    • JL: Is Jenna still thinking about estimating abundance? JE: paper is in press, could use her data to test the corrections
      • KP: Good idea to test the method on this or some other gold standard data set or simulation

• Update from Guillaume
  • Generating new protein family HMMs
    • HMM search of Dongying's 100 families (universal or big, but not markers per se) vs. IMG
    • BLAST all vs. all on all sequences that didn't hit a family
    • MCL on these BLAST hits to cluster them into families
    • Just finished 20 day process of generating these results
      • 350,000 families (6,500 with >75 members)
      • MCL may need to be fine-tuned
    • Next: will build HMMs for these new protein families that we can use in various analyses
      • Will search back against data sets to see if the new HMMs look robust
    • Morgan will look at clusters that don't hit any previously described family (or in PFAM but no known function)
    • Tom will subdivide existing families into subfamilies, looking for novel subfamilies in metagenomic data
      • This subfamily approach could be applied to Morgan's families also
    • Goal: data freeze in the next month or so

• Dongying: What deep coverage data sets are out there besides GOS
  • JE: Bejing's human gut data is available in the short read archive at NCBI (maybe BioTorrents - coming soon), much on their website

• Steve: Bejing gut paper assembled Illumina reads into contigs - what do we think about that?
  • TS: used Sanger scaffolds to do some testing of the process
  • JE: checked some against known assemblies
  • AD: the Bejing assembler has not been documented
    • Variable coverage breaks most assemblers
  • JE: generally not a good idea to use assembled reads for metagenomic analyses (because they generate hybrids)
    • OK to leverage contigs/alignments in population level analyses
    • Tile reads vs. a reference assembly if you want to compare them to each other
    • Best not to use the consensus for downstream
  • JE: first Illumina based study that looks good - Illumina might be more promising for metagenomics than originally thought
    • Uses much less DNA (~1 microgram) than 454 (20-40 micrograms, though this may be coming down)
    • Need to check diversity measures vs. other data sets
    • This genome-guided approach get better as more human microbiome genomes are sequenced
  • Data here http://gutmeta.genomics.org.cn/