Moore Notes 3 23 10
From OpenWetWare
Jump to navigationJump to search
Group Call
- Quarterly report (due end of March)
- a few bullet-points each, not more than 1 paragraph
- what have you done in March? (since annual report)
- post to wiki by Friday
- Update from Alex Koeppel: Slides
- observes a curved "linear" section, and no flare when applying ecotype method to GOS data
- ecotype method has problems with missing data (due to non-overlapping, incomplete GOS reads)
- groups of reads that align to a section of the gene will cluster together in the tree
- method assumes that there is one rate of ecotype gain across the whole tree - might not be true if sequences are divergent
- Josh: why not use branch lengths directly, rather than binning? Alex: Wesleyan group is working on this
- Katie: try a model that accounts for multiple substitutions to estimate rate of sequence change, rather than percent identity
- Tom: the OTU group is using phylogenetic distances to handle non-overlapping reads, which might help here
- try not to use Neighbor-joining phylogenetic tree methods
- try to use a phylogenetic method that handles missing data well
- e.g., FastTree or RAxML with iterative placement
- Sam: if RAxML is slow, try it with the pruned down alignment that doesn't have too much missing data just to see if branch length works better than the binning criteria. Martin: not sure we want to work on changing the ecotype simulation software.
- Martin/Alex: 16S curve looks OK, probably because it is slower evolving. Katie: try focusing on conserved sections of proteins, and work with amino acid sequences.
- Tom: complete linkage binning over-estimates the number of OTUs. Might want to try average linkage if possible.
- Martin: different marker genes are useful for different clades since clades differ in their rates of evolution. Sam/Josh: Better methods will help.
- Katie: What about using Dongying's clade-specific markers (e.g. or a version based on a slow rate of evolution in that clade)?
- Alex: another program to try is adaptML, although it doesn't account for periodic selection