<html>
I have just run limma and am getting p-values after eBayes that are
smaller than the p-values before, leading to 100% of my genes being
declared significant at any value of FDR you care to use.
<br><br>
The design is a 1-way ANOVA with 6 treatments and 2 reps/treatment (which
I know is not great but ...)<br><br>
I thought that the denominator adjustment would make the posterior
sigma^2 > unadjusted MSE, but this is not the case. Here
are the commands I used to fit the model and do the ebayes
adjustment. <br><br>
<font face="Arial, Helvetica" color="#800080">contrast.matrix<br><br>
</font> c1 c2 c3 c4 c5<br>
trt1 1 1 1 1 1<br>
trt2 -1 1 1 1 1<br>
trt3 0 -2 1 1 1<br>
trt4 0 0 -3 1 1<br>
trt5 0 0 0 -5 1<br>
trt6 0 0 0 0 -5<br><br>
<font face="Arial, Helvetica" color="#800080">fitRMA=lmFit(RMAdata,contrast.matrix)<br>
fitAdj=eBayes(fitRMA)<br><br>
</font>ls.print(lsfit(fitRMA$sigma^2,fitAdj$s2.post))<br>
Residual Standard Error=0<br>
R-Square=1<br>
F-statistic (df=1, 22744)=1.632754e+35<br>
p-value=0<br><br>
Estimate
Std.Err t-value Pr(>|t|)<br>
Intercept 0.0093 0
1.026963e+17 0<br>
X
0.5628 0
4.040735e+17 0<br><br>
<font color="#800080">mean(fitAdj$s2.post)<br>
</font>[1] 0.02991697<br><br>
<font color="#800080">mean(fitRMA$sigma^2)<br>
</font>[1] 0.03656270<br><br>
<font color="#800080">fitAdj$s2.prior<br>
</font>[1] 0.02136298<br><br>
<br>
<x-sigsep><p></x-sigsep>
Naomi S.
Altman
814-865-3791 (voice) <br>
Associate Professor <br>
Bioinformatics Consulting Center<br>
Dept. of
Statistics
814-863-7114 (fax) <br>
Penn State
University
814-865-1348 (Statistics) <br>
University Park, PA 16802-2111<br>
</html>