phylocom, nti and nri and p-values

[Sonja]

Hi
I had some questions about the nti and nri values with the
relatedness.py script. I have used whole tree (otus from 6 different
samples) as input and, as it says on the script description, provided
the list of the otus in each particular sample I want to analyze. I
get very high nti values (7-10) and nri values ranging from 0.7 to 7.
However, how do I know if these results are significant if I dont have
any p-values? Also, how are the randomizations done?
I would like to be able to say if my samples are phylogenetically
clustered or not, but to me it looks like I can only compare my nri
values and say "sample x is more phylogenetically clustered than
sample y". For the NTI values, they seem very high...

Thanks!

Cheers, Sonja

[Dave Vuono]

That makes sense that NTI is higher. Think about what NTI is actually
doing (nearest taxon) rather than the NRI which looks at the whole
tree. Since the taxonomic resolution of our sequences is typically not
uniform (e.g., not all up to the genus level), the NTI gets tripped up
and the nearest taxon may be from a different taxanomic rank. At least
thats the best explaination I can come up with... Thoughts anyone? As
for p-values... Depends on your question, although the best way would
be to have enough replicates and do a parametric test. I would suggest
reading Webb's 2000 and 2002 papers for more details.

[Will Van Treuren]

Hi Sonja,

I haven't yet implemented calculations for the p-values like Phylocom,
but I hope to make those available in a few weeks in the development
version of Qiime. As far as your values being high, I am not sure
thats a problem. Dave has a good suggestion, but I don't really know
what values to expect. How big is your tree, and what are the branch
lengths (if any) like? One thing that may be contributing to higher
than expected scores is formula differences. Webb gives subtly
different formulas for NRI/NTI in 3 different places: Webb 2000, Webb
2002, and the Phylocom manual. I implemented the formula as given in
the Phylocom manual.
As far as the randomization procedure, I used null model 2 from
Phylocom, which is random selection of taxa without replacement from
every taxa on the tree. I hope to implement the other models the
Phylocom manual discusses in the next couple weeks.

Will

[MikeyJ]

Hi Sonja,

I believe positive NRI and NTI values indicate clustering and negative
would indicate overdispersion, so you have a certain amount of
clustering in all your samples (which I would expect would generally
be true of most microbial communities). That suggests that the
communities are under some sort of selection, which is useful
information, if not particularly ground-breaking or interesting. How
I usually go on to look at the data is to see whether NTI or NRI (or
both) correlate significantly with any of the other metadata,
particularly continuous data that I have. If you get any strong
associations it gives you an idea of the most important variables for
forcing community structure (though, it's association, not causation),
which, if you're lucky, can be interesting :)

I like the NRI and NTI metrics more than diversity indices like
Shannon or Simpson's because there is this implied selection, whereas
Shannon and Simpson are more like summary statistics for community
structure.

My take on it is probably abbreviated and wrong though - any proper
ecologists out there for comment?

Cheers

Mike

[JJ. Wang]

Hi Sonja,

NTI/ses.mntd may be more appealing for microbial communities. Here are two
recent references with 454 method. Hope it helps.

Wang, J., J. Soininen, J. He, and J. Shen. 2012. Phylogenetic clustering
increases with elevation for microbes. Environmental Microbiology Reports
4:217-226.

Stegen, J. C., X. Lin, A. E. Konopka, and J. K. Fredrickson. 2012.
Stochastic and deterministic assembly processes in subsurface microbial
communities. The ISME Journal. Online

Cheers,
Jianjun

--------------------------------------------------
From: "MikeyJ" <mikey...@gmail.com>
Sent: Wednesday, May 30, 2012 3:55 PM
To: "Qiime Forum" <qiime...@googlegroups.com>
Subject: [qiime-forum 1.4.0] Re: phylocom, nti and nri and p-values