Entering edit mode
Jain, Nitin
▴
70
@jain-nitin-1028
Last seen 10.0 years ago
Dear Lisa,
The function 'fdr.adjust' uses for loops in R, which are not very
efficient. However, I have tried it on 64-bit Linux machine in
Pfizer, and it runs quite fast (< 1 hr) for 1000-iterations.
LPE is best suited if the number of replicates and samples are less
than or equal to 5. We upgraded the LPE package from the original
paper (2003), and published a paper in BMC Bioinformatics. The latest
version is on Bioconductor since early last year.
(http://www.biomedcentral.com/1471-2105/6/187)
Originally, when we collaborated with S-Plus, then there was not the
comprehensive 'resampling' based fdr adjustment. I am not aware about
whether S-Plus function has been upgraded or not. Also, am not sure if
they are using the C-code for the resampling part or the regular
S-engine. I am cc'ing this mail to Dr. Lee in Univ. of Virginia, who
can shed more light on this.
To obtain Bonferroni adjusted values, just run lpe test (without
fdr.adjust) and obtain "median.diff" and "z.stats" (these values are
in the result of lpe function). Then calculate the p-values for each
gene:
pnorm.diff <- pnorm(median.diff, mean = 0, sd = std.dev)
p.out <- 2 * apply(cbind(pnorm.diff, 1 - pnorm.diff), 1, min)
and multiply the p-values (p.out) by the number of genes to obtain the
Bonferroni adjusted p-values (use pvalue =1, if it is greater than 1).
We will add this option (Bonferroni) in the new version of LPE
package.
Best,
Nitin
On 2/15/06, Lisa Croner <lisa.croner at="" biogenidec.com=""> wrote:
>
> Hello Dr. Jain,
> I have been using your LPE Bioconductor package (in R), and find
that fdr.adjust with resamp or mix.all is very slow on both my windows
and new linux machines. It appears that a 1000-iteration run may take
16 hours to complete. On the other hand, a colleague tells me that
running fdr.adjust with resamp for 1000 iterations under S-Plus
completes in about 5 minutes on his (windows, I believe) laptop. Does
this make sense to you? Is it know that S-Plus is so much faster than
R? Do you perhaps have a faster version of LPE to run under R?
> One other question: How can one run simply the Bonferroni adjusted
LPE test from your 2003 Bioinformatics paper (Figure 2 and page 1948),
using the LPE R package?
> Many thanks for further information,
> Lisa
>
> --------------------------------------------------------
> Lisa J. Croner, Ph.D.
> Scientist II, Computational Biology
> Biogen Idec
> San Diego, CA
> (858) 401-5669
> lisa.croner at biogenidec.com
> --------------------------------------------------------
>
______________________
Nitin Jain, PhD
<nitin.jain at="" pfizer.com="">
Non Clinical Statistics
Pfizer, Inc. (Groton, CT)
Bldg: 260, # 1451
Ph: (860) 686-2526 (Office)
Fax: (860) 686-6170
----------------------------------------------------------------------
LEGAL NOTICE\ Unless expressly stated otherwise, this
messag...{{dropped}}