Moore Notes 10 2 13

Group Call

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

• Guillaume: Jenna has been working with PICRUSTs and QIIME
  • She will be helping Guillaume run analyses

• Josh: Tara Oceans data
  • Shotgun metagenomics samples from Pacific and Atlantic
    • 243 samples, 68 locations
    • 5m depth
    • 40-450 million reads per sample
  • Will be available at the end of November
  • Katie: Any gaps in their sampling locations?
    • Josh: hard to find out, well distributed, some bias towards Mediterranean
    • Might have to wait until released to figure this out
  • Should get in touch with senior PIs: Bork, Acinas, Hingamp, Raes, Fallows, Sullivan
    • Tom will drop Sullivan a line copying Josh, Jonathan, Katie

• Sarah: phylogeography background, has focused on species in vertebrate studies
  • Phylogeography looks at distribution of variants
  • How phylogenetic distribution of genetic variants scales across space and time
  • For bacteria, she would do phylogeography of function using SFams database
  • Blast samples versus SFams, focusing on ones where we know functions (e.g., photorhodopsin)
  • Tom: concerned about having enough resolution in a metagenomic read to do this
  • Katie: look at Sam's paper, maybe do some additional simulations
  • Maybe looking at reads that cluster together in tree would increase confidence
  • Tom: might be good to work with nucleotide sequences (for phylogeny, not for read classification)
  • Needs to think about how to address different sources of variants, in a family specific way, in order to set expectations/null distributions
  • How to synergize with Josh?
    • Use same ShotMap runs (e.g., on Tara Oceans)
    • Niche modeling plus phylogeography
      • To predict distributions of variants, which might be more important functionally than distributions of families
      • Example: nitrate reductase subclade that alters its enzymatic reaction (and output of the metabolic pathway)

• Ladan: Predicting functions of SFams with no GO annotations
  • Mapped SFams into a network (weighted, nondirected edges) based on Pearson correlation of presence-absence across genomes
  • Tried to find tightly connected subnetworks with extreme sets algorithm
    • Finds groups of nodes where removing a piece has a cost greater than removing the whole group
    • There may be extreme sets within extreme sets (hierarchical, tree-like structure)
  • Could potentially map functions from annotated members of the extreme set
    • Most sets are nearly all annotated or mostly not annotated
  • Stephen: how does the phylogenetic distribution of the genomes impact the results?
  • More info here: https://docs.google.com/file/d/0B5MwVN20vJJzZjJJZW9nOFIzems/edit?usp=sharing
    • Dongying: tried a "phylogenetically independent correlation" (from the 80s), but didn't make sense and eliminated some good information
  • Stephen: could evaluated results based on known annotations
    • Ladan: tried this to compare her algorithm to MCL