Suspicious results from CuffDiff

[Pejman]

Hey folks
I'm running CuffDiff from the release 0.9.2, for RNAseq. I use it with tophat, and Reference annotation for human data. Here's the simple running command:

cuffdiff --FDR 0.01 -N hg18_annotation01.gtf accepted_hits1.bam accepted_hits2.bam

There are few points of concern

first: over 90% of the genes with "status = OK" have "Significant = yes" which is kinda not so nice.

second: that there are so many genes with very low FPKM that are among the significant ones, which can also lead to fold changes like: 1.7977e+308 !

and third: There are genes with large FPKMs that have not been tested (status = NOTEST) , here's some sample, from the file "genes_expr.diff", with a "NOTEST gene with hight FPKM, and two low FPKM genes with test status "OK":

ttest_id	status	value_1	value_2	ln(fold_change)	test_stat	p_value	significant
NR_036659	NOTEST	1119.85	6601.74	1.774 		-54.911 	0	no
NM_001130440	OK	5.6E-09	5.7E-09	0.030 		-1.744 		0.081	no
NM_001001977	OK	9.7E-09	1.2E-08	0.227	 	-18.416 	0	yes

Does somebody have some idea? :-/

thanks
pej

[Boel]

Hi Pejman,

Any progress with the fact that so many of the genes with "status = OK" have "Significant = yes"? Kind of seeing the same thing.

Boel