Moore Notes 12 4 13

Group Call

• Participants: Katie, Jonathan, Tom, Stephen, Guillaume, Dongying, Ladan, Patrick

• Jonathan: Microbial ecology in N Atlantic
  • Healy Hamilton (ex-Berkeley/Cal Academy)
  • Now at foundation that specializes in biodiversity databases
  • NOAA call for proposals for biodiversity databases in N Atlantic
  • She will try to convince NOAA to collect microbial samples on the N Atlantic ships

• Tom/Guillaume: Talked to Morgan about PICRUSTs
  • Thinks that PICRUSTs failures are probably due to reference database gaps
  • Compare Shotmap vs. MG-RAST (SEED converted to KEGG) to confirm that Shotmap abundances agree with MG-RAST
  • Will collapse KEGG OGs into pathways
  • Will also check the software version

• Josh: Any news on Tara Oceans?

• Stephen: Relative abundance of plasmids across samples
  • Could they explain differences in average genome coverage?
    • JE: Yes, because copy number of plasmids can be very high (varies a lot)
  • JE: Very hard to identify plasmids, though there are a couple gene markers
    • Chromosome partitioning factors for plasmids (parA, parB) tend to be phylogenetically clustered, so can be identified
    • In some cases, there are genes that consistently occur on plasmids
    • Some people look for assemblies that circularize (need to be high coverage)
  • SN: What about mapping reads to plasmids in IMG?
    • JE: Thinks HMP did this
    • JE: Similar to problem of finding free phage DNA

• Patrick: Correcting functional signals for taxonomic abundance
  • To what extent are differences in gene family abundance driven by taxonomic variation?
  • JE: PICRUSTs gives the expected distribution of functional genes for a taxonomic arrangement

• Ladan: slides
  • Extreme set clustering of halophile protein families
  • Predicting GO annotations of genes based on clustering with annotated genes
  • Comparison of extreme sets vs. MCL, with two different rules for assigning GO to nodes within a cluster
  • How to determine precision and recall?
    • KP: Treat each annotation of a GO term to a gene (yes vs. no) as an independent test
    • TS/KP: Could partition genes based on characteristics (e.g., number of GO terms, extreme set properties) and check if performance differs across partitions
  • Stephen: Try extreme sets on a benchmark where MCL has already been validated