Metabolomics Society Forum

Software => Other => Topic started by: the_august on November 01, 2014, 05:00:03 AM

Title: Metabolite PLS-DA analysis bad p-value
Post by: the_august on November 01, 2014, 05:00:03 AM
Hi All,

I'm doing some metabolomics analysis using the web based tool available from http://www.metaboanalyst.ca/ (http://www.metaboanalyst.ca/). I'm using a concentration table as input.

For the PLS-DA analysis my samples good separtion, however the permutation p-value is bad. I'm very new to this kind of analysis, and I'm not sure if I'm making some mistakes. Can permutation values be bad if separation between samples are good? Moreover as the PLS-DA model is bad, can I use the metabolites with VIP values >1.0? Or are these VIP values useless as well?

(http://http://s10.postimg.org/bjceqfca1/PLS_DA_combined.png)

For bigger sized image please click on the following link:

http://postimg.org/image/s73wsx71h/ (http://postimg.org/image/s73wsx71h/)

Thanks
Title: Re: Metabolite PLS-DA analysis bad p-value
Post by: sneumann on November 04, 2014, 01:38:00 AM
Hi august,

I don't have the first hand knowledge on PLS-DA to comment on your question,
but I remember the excellent keynote by Rasmus Bro at Metabolomics 2012
( http://www.metabolomics2012.org/keynote-speakers/ (http://www.metabolomics2012.org/keynote-speakers/) ) where he used
Matlab (his favourite tool, but can probably done in R as well) to draw
a fully RANDOM dataset, a fully RANDOM Y and got a (visually) nice separation
of the classes.

Maybe this Metabolomics journal article can also help ?
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3337399/ (http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3337399/)

Yours,
Steffen

P.S. : If you succeed to reproduce Rasmus' PLS-DA on random data in R,
I'd love to see a code example posted here...