Entering edit mode
8.5 years ago
tonja.r
▴
600
I am using edgeR for my analysis and I have found out that if I break a function estimateDisp(y,design) into two sub functions (and it is stated in manual that it is the same):
estimateCommonDisp(y,design)
estimateTagwiseDisp(y_two)
I get a bit different FDR and Pvalues:
cds =DGEList(counts = counts(set),group=cond)
y <- calcNormFactors(cds, method="upperquartile")
y_two=estimateCommonDisp(y,design)
y_two=estimateTagwiseDisp(y_two)
fit_empirical<-glmFit(y_two,design)
lrt_empirical<- glmLRT(fit_empirical, coef=2)
first_top<- topTags(lrt_empirical, n=nrow(set))$table
head(first_top)
logFC logCPM LR PValue FDR
ENSMUSG00000022066.8/Entpd4 -0.6415939 6.271973 28.34501 1.015083e-07 0.00138658
ENSMUSG00000034248.6/Slc25a37 -0.6227765 7.262966 27.89511 1.280728e-07 0.00138658
ENSMUSG00000028228.5/Cpne3 0.5208529 6.618665 19.04765 1.274949e-05 0.09202154
ENSMUSG00000028229.5/Fam82b 0.4839051 6.248222 15.75029 7.227703e-05 0.39125363
ENSMUSG00000041058.8/Wwp1 0.4560114 6.541054 14.56492 1.354118e-04 0.58641438
ENSMUSG00000073987.3/Ggh 0.4459020 6.039896 13.53324 2.343752e-04 0.84582099
Just estimateDisp
:
cds =DGEList(counts = counts(set),group=cond)
y <- calcNormFactors(cds, method="upperquartile")
y_one <- estimateDisp(y,design)
fit_empirical<-glmFit(y_one,design)
lrt_empirical<- glmLRT(fit_empirical, coef=2)
first_2_top<- topTags(lrt_empirical, n=nrow(set))$table
head(first_2_top)
logFC logCPM LR PValue FDR
ENSMUSG00000034248.6/Slc25a37 -0.6221628 7.262583 48.74556 2.914180e-12 6.310074e-08
ENSMUSG00000022066.8/Entpd4 -0.6398572 6.270966 43.89002 3.473554e-11 3.760643e-07
ENSMUSG00000028228.5/Cpne3 0.5208468 6.619020 19.90427 8.141830e-06 5.876502e-02
ENSMUSG00000073987.3/Ggh 0.4459022 6.040336 15.88664 6.725166e-05 2.911755e-01
ENSMUSG00000028229.5/Fam82b 0.4838955 6.248296 15.71660 7.357573e-05 2.911755e-01
ENSMUSG00000041058.8/Wwp1 0.4560675 6.541315 15.54219 8.068412e-05 2.911755e-01