Entering edit mode
Guest User
★
13k
@guest-user-4897
Last seen 9.6 years ago
Hi everybody,
I have a question regarding comparing results from a t-test and limma.
I compared the p values obtained from both algorithms using 100
samples in each condition. Given the large number of observations my
expectation was to see high correlation between the p values. As shown
below, I ran the same code in three different conditions for the mean
and standard deviation. The mean and standard deviation loaded from
the google docs files are from a real micro array experiment.
1. mean and std from microarray exeperiment: no correlation between p
values
2. constant fold change and std from microarray exeperiment: very
small correlation
3. constant fold change and uniform std from 0.01 to 0.2: high
correlation
I understand that limma uses information across genes, but shouldn't
this information be weighed with the number of observations for each
condition?
I put the source code here, since on the mailing list backslashes
disappear.
https://drive.google.com/file/d/0B__nP63GoFhMZEFUbjNYTlFJWm8/edit?usp=
sharing
Thank you,
Giovanni
-- output of sessionInfo():
R version 2.15.2 (2012-10-26)
Platform: x86_64-w64-mingw32/x64 (64-bit)
locale:
[1] LC_COLLATE=English_United States.1252
[2] LC_CTYPE=English_United States.1252
[3] LC_MONETARY=English_United States.1252
[4] LC_NUMERIC=C
[5] LC_TIME=English_United States.1252
attached base packages:
[1] grDevices datasets tcltk splines graphics utils stats
[8] grid methods base
other attached packages:
[1] limma_3.14.4 genefilter_1.40.0 Biobase_2.18.0
BiocGenerics_0.4.0
[5] RCurl_1.95-4.1 bitops_1.0-6 reshape2_1.2.2
Hmisc_3.14-3
[9] Formula_1.1-1 survival_2.37-7 lattice_0.20-29
loaded via a namespace (and not attached):
[1] annotate_1.36.0 AnnotationDbi_1.20.7 cluster_1.15.2
[4] DBI_0.2-7 IRanges_1.16.6 latticeExtra_0.6-26
[7] parallel_2.15.2 plyr_1.8 RColorBrewer_1.0-5
[10] RSQLite_0.11.4 stats4_2.15.2 stringr_0.6.2
[13] XML_3.98-1.1 xtable_1.7-3
--
Sent via the guest posting facility at bioconductor.org.