Re: Questions about LEfSe

[Curtis Huttenhower]

Hi Luc - thanks for getting in touch; I've CCed in the LEfSe users group here, and especially Siyuan and Sagun on our end for more info. My quick first-pass response is that 1) in the absence of subclasses, the test is basically just a K-W (since the Wilcoxon is for consistency across subclasses), and 2) a "good" sample size will depend on the distribution of each feature, since it's easier to get an estimate for a prevalent bug (e.g. Fprauz or most Bacteroides in the gut) from just a few samples than for a bug that's often absent (e.g. Pcopri). So if you want ~5-10 nonzero measurements to get a decent estimate of the direction of effect, that only takes 5-10 samples for an abundant bug, but could easily take 50+ for a rare one.

Thanks again -

Curtis

On Fri, Nov 29, 2019 at 5:00 AM Luc Dantan <luc.d...@univ-perp.fr> wrote:

Hello,
I am currently a PhD student and your LEfSe tool seemed to be a really
relevant tool for my S16 sequencing data analysis.
Nevertheless, I would like to ask you a few questions:
First, it seemed to me that the first step was a Kruskal-Wallis test on
classes and the second step was a Wilcoxon test on subclasses. However,
if I don't have a subclass in my data, will it distort my analysis? If
so, is there any parameter to change to avoid this problem?
Second, again in relation to the Wilcoxon test in step two, you say in
your article (on page 13) that if the subclass has few samples, it may
reduce the ability to identify biomarkers. In your opinion, how many
samples at least are needed to avoid this problem?
Thank you in advance for your time and consideration.

Sincerely,

Luc Dantan


--
Luc DANTAN
Doctorant / PhD student
IHPE - UMR5244
58 Avenue Paul Alduy, Bât R
66860 Perpignan Cedex - France
http://ihpe.univ-perp.fr