Strange NA values

[Jean Keller]

Dear EBSeq community,

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