Dear all,

I am using EBSeq on RSEM output in a context of differentially expressed genes detection. I have 2 conditions (C1, C2) and three replicates in each. 

When using EBSeq in R (last versions of all programs), I got around 2000 genes with the NA status. However, when I used EBSeq in RSEM (rsem-run-ebseq), these genes were not anymore NA but all considered as DE at a FDR threshold of 0.01.

Following the EBSeq's manual, I tried to change the ApproxVal in the EBTest function but it did not improve anything. I tried to have look to the normalized counts of these genes through the function GetNormalizedMat and subset of results are as follow:

Gene_ID    EBSeq_inR    PPDE_rsem-run-ebseq    C1    C1.1    C1.2    C2    C2.1    C2.2

Cluster-10257.0    NA    0,99999998    8,725090051    10,99097949    19,72022482    0    0,739891811    0

Cluster-106155.2    NA    0,999806082    444,3937651    545,0884685    521,9791817    323,3767238    390,0857608    372,08312

Cluster-34655.0    NA    0,999977486    0    0    0    47,88561335    0    0

Cluster-2228.0    NA    1    19,94306297    31,14110854    78,8808993    0    0    0

Cluster-37095.0    NA    0,999886204    766,4866966    669,5338337    748,3218545    534,0410674    568,9768029    449,0450628

I understand why Clusters 34655.0, 2228.0 and 10257.0 got a NA status under R, however I can't explain why the others (Cluster-106155.2, Cluster-2228.0, Cluster-37095.0) got a NA status (except maybe for the Cluster-2228.0 which have three 0 for one conditions). 

I compare my script with the one implemented in RSEM and both were identical. 

Could someone help me to understand this ?

Thank you,

Jean