methods for fetching unique/rare taxa

[wasade]

New methods for determining rare and unique taxa from a BIOM TaxonTable (e.g., from summarize_taxa.py). New tree objects were required since cogent is GPL, but this fell together pretty well.

@JWDebelius, I think this should be good to go for your uses. The tables are optionally filtered, as filtering out the rare/unique taxa from a BIOM table does add a fair amount of overhead. See the main block for an example of how to use the code, and of course do not hesitate to send questions.

@gregcaporaso @rob-knight, this may be generally useful for QIIME as well. It's threshold based, and is not doing any stats, but the framework is a much more natural way to deal with taxonomy in these datasets. The tree could easily be further annotated as well. The central idea, given some tax strings:

tax_strings = ["k__foo; p__bar; c__123",
                     "k__foo; p__bar; c__",
                     "k__foo; p__bar; c__456",
                     "k__foo; p__other; c__789"

The method constructs the tree:

(((c__123,c__456)p__bar,(c__789)p__other)k__foo)root"

Taking this a little further, the method accepts taxon strings per sample, which allows you to annotate the tree to determine how many samples a particular node was observed in. It handles unclassified samples cleanly as well. It does not currently check for contested Greengenes groups, though that could be added easily.

[rob-knight]

Yes I think this will be generally useful. Thanks!

On Nov 6, 2013, at 8:31 PM, Daniel McDonald notifications@github.com<mailto:notifications@github.com> wrote:

New methods for determining rare and unique taxa from a BIOM TaxonTable (e.g., from summarize_taxa.py). New tree objects were required since cogent is GPL, but this fell together pretty well.

@JWDebeliushttps://github.com/JWDebelius, I think this should be good to go for your uses. The tables are optionally filtered, as filtering out the rare/unique taxa from a BIOM table does add a fair amount of overhead. See the main block for an example of how to use the code, and of course do not hesitate to send questions.

@gregcaporasohttps://github.com/gregcaporaso @rob-knighthttps://github.com/rob-knight, this may be generally useful for QIIME as well. It's threshold based, and is not doing any stats, but the framework is a much more natural way to deal with taxonomy in these datasets. The tree could easily be further annotated as well. The central idea, given some tax strings:

tax_strings = ["kfoo; pbar; c123", "kfoo; pbar; c", "kfoo; pbar; c456", "kfoo; pother; c789"

The method constructs the tree:

(((c123,c456)pbar,(c789)pother)kfoo)root"

Taking this a little further, the method accepts taxon strings per sample, which allows you to annotate the tree to determine how many samples a particular node was observed in. It handles unclassified samples cleanly as well. It does not currently check for contested Greengenes groups, though that could be added easily.

---

You can merge this Pull Request by running

git pull https://github.com/wasade/American-Gut taxtree

Or view, comment on, or merge it at:

https://github.com/qiime/American-Gut/pull/26

Commit Summary

• methods for fetching unique/rare taxa

File Changes

• A americangut/taxtree.pyhttps://github.com/qiime/American-Gut/pull/26/files#diff-0 (152)
• A tests/test_taxtree.pyhttps://github.com/qiime/American-Gut/pull/26/files#diff-1 (167)

Patch Links:

• https://github.com/qiime/American-Gut/pull/26.patch
• https://github.com/qiime/American-Gut/pull/26.diff

[wasade]

@adamrp @ElDeveloper Can you review please and merge if sane?

[wasade]

Thanks for comments, will address. Not sure if its appropriate right now to have a script interface, the use of argv is just as a simplistic example. But, can gut if you think it'd be good to do so

[adamrp]

Thanks Daniel, I think this looks really good. Minor comments mostly. Not sure of how best to get around that list lookup, although I do think it could end up bing a time-sink for very large trees.

[wasade]

Only for nodes with a large number of children, which of course could happen but should be niche. Set/dict would outperform once there were more than around 3-5 children. This will happen in taxonomy but the hierarchies are pretty small. For fasttree and phylogeny, the vast bulk of the nodes should be bifurcating

On Thu, Nov 7, 2013 at 1:53 PM, adamrp notifications@github.com wrote:

Thanks Daniel, I think this looks really good. Minor comments mostly. Not sure of how best to get around that list lookup, although I do think it could end up bing a time-sink for very large trees.

— Reply to this email directly or view it on GitHubhttps://github.com/qiime/American-Gut/pull/26#issuecomment-28005347 .

[wasade]

@ElDeveloper @adamrp safe to merge?

[ElDeveloper]

:+1:

[wasade]

thanks

On Fri, Nov 8, 2013 at 11:09 AM, Yoshiki Vázquez Baeza < notifications@github.com> wrote:

[image: :+1:]

— Reply to this email directly or view it on GitHubhttps://github.com/qiime/American-Gut/pull/26#issuecomment-28084241 .