Moore Notes 12 5 12

Group Call

• Participants: Stephen, Guillaume, Katie, Jonathan, Josh

• Regular call time for next quarter: 10am Wednesdays

• Data from MMI
  • Ian Joint from AMT cruises (ian.joint@gmail.com)
  • Working on contact at Tara (could be hard)
  • Do we want to be connected with MEGX? (maybe Nicole) - YES

• GBMF investigators announced

• Guillaume is working on automatic updating code
  • Will be ready to run when Tom is back in Jan

• Update from Stephen
  • Recall is low for SFams with large family size
    • Common in genomes
    • Also more diverse and more connected in family network
  • LAST performs well and is fast (using non-stringent parameters)
    • Expanded simulation study to simulate 100 reads from every SFam to get family level performance statistics
    • About 12% of SFams have 0% recall (when doing leave-one-out analysis, better if perfect match is there)
      • For some due to strong similarity with a nearby SFam
      • For some due to low homology within SFam
    • Multiple best hits (ties) is probably playing a role
    • Similarity of consensus sequences does not predict misclassification of reads
    • Looking at all-v-all BLAST results to see if nearest neighbor sequence is a similar distance for sequences in own vs. other SFam
  • Will look at TIGRFAMs to see if results are specific to SFams or are also found with other HMM-based family dbs

• JE: if/when we build new families, we should think about what genomes go into the pipeline
  • Should we include fragments from metagenomes
  • What is the seed set?
  • GBMF investigators might have some single-cell genomes and other new genomes that are not in dbs
  • Metaproteomics data?

• Update from Josh
  • Working on code to do beta-diversity niche mapping
  • Here predicting community similarity as a function of distance between samples
  • Can make maps, e.g.,
    • Classify regions into ecological types (eco-regions)
    • Visualize community turnover in space
  • Well-developed methods for macroorganisms can be applied to microorganisms pretty directly
  • Could do cross-validation on marine macroorganism data (from grid cells where we have metagenomics data) to test the idea
  • Model selection is computationally intensive - modifying to get it working on the cluster