[BioC] EBSeq Results Affected By Outlying Values

[Dario Strbenac]

Hello,

I find that some transcripts found by EBSeq have a problematic pattern in the raw data, where only one sample has any counts. For example, in a 3 vs. 3 scenario, these are some of the transcripts which have posterior probability of D.E. = 1 :

                    Sample13 Sample14 Sample15 Sample16 Sample17 Sample18
comp209187_c0_seq13     0.00   112.99   107.04     0.00     0.00     0.00
comp209188_c11_seq1    40.00    38.91    42.89   107.00   155.96   105.58
comp209189_c8_seq1    248.02   235.06   287.10    18.82    93.13    10.74
comp5876_c0_seq1        0.00     0.00     0.00   128.00     0.00     0.00
comp701657_c0_seq1      0.00     0.00     0.00   187.00     0.00     0.00
comp85095_c0_seq1       0.00     0.00     0.00   125.01     3.00     0.00

The first 3 samples are in one condition, the last 3 are in the other condition.

The values are expected counts from RSEM, which the journal article says EBSeq works with. I'm a bit cautious about claiming the last 3 transcripts are D.E. Is there anything in EBSeq analogous to the new EdgeR-Robust algorithm which can deal with these observations ?

--------------------------------------
Dario Strbenac
PhD Student
University of Sydney
Camperdown NSW 2050
Australia