Zero counts in one condition causing no tests and false negatives, and ApproxVal parameter setting has no effect on outcome

[roberthaney]

######In running EBSeq_1.10.0 with two treatments, with 2 replicates each, I get NA values for differential expression for a number of transcripts from multiple transcript genes that have large fold change differences between treatments T1 and T2, see counts and FC below (csv format) for example transcripts.

transcript_name,T1a_Count,T1b_count,T2a_count,T2b_count,PostFC
MSTRG.32727.1,556609,351716,0,0,708006
MSTRG.32727.2,391045,210838,2,0,333938
MSTRG.32970.5,15994,7871,0,0,17999
MSTRG.32970.7,31916,18305,0,0,38643
MSTRG.32970.6,18322,9736,0,0,21374
MSTRG.32970.1,44568,24467,0,0,52813
MSTRG.32970.3,10950,6447,0,0,13434
aug3.g25588.t1,15684,8038,0,0,17986
MSTRG.6065.1,1498,448,0,0,1383
MSTRG.15150.1,14834,7583,0,0,16991
aug3.g26326.t1,1162,350,0,0,1076
MSTRG.2214.2,67402,29329,0,0,71818
MSTRG.2214.3,28283,15046,0,0,33011
MSTRG.2214.6,6148,2390,0,0,6252
MSTRG.2214.7,3337,1101,0,0,3188
MSTRG.2214.8,5337,2375,0,0,5743
MSTRG.2214.9,9320,3218,0,0,9052

######For other transcripts, I get a PPDE of 0 and PPEE of 1 with the following data:

transcript_name,T1a_Count,T1b_count,T2a_count,T2b_count,PostFC
MSTRG.2214.11,12240,16191,0,0,24446
MSTRG.536.2,7608,599,0,0,5261
MSTRG.15150.2,11726,1308,0,0,8513

#####Yet other transcripts at the same genes are called DE, often with less-biased expression, as long as they did not have zero counts in both replicates of one treatment.

######In the case of gene MSTRG.2214, several other transcripts are called as DE with the following counts and PostFC, yet some of the transcripts described at this locus above are more biased in their expression towards one treatment:

MSTRG.2214.1,0,3,448,218,0.031
MSTRG.2214.4,32632,14091,0,3,21110
MSTRG.2214.5,23926,15318,0,3,18666

######In the case of gene MSTRG.32727, a third transcript at this locus is called as DE with the following less-biased expression data:

aug3.g10236.t1,6029,2592,6,7,1763.72

######Another transcript at locus MSTRG.6065 is called DE with the following data, and it is less biased in expression than the transcript called as NA above:

MSTRG.6065.3,1326,846,27,23,154

######At locus MSTRG.536, another transcript is called DE with the following counts, and a postFC quite similar to the transcript called as EE(MSTRG.536.2):

MSTRG.536.1,23985,12708,0,20,5293

######In the case of gene MSTRG.32970, while the gene is called as DE, none of the transcripts are, this is also true of gene MSTRG.15150, which is DE, but none of its transcripts are, although they are highly biased in expression towards one treatment.

######Library size parameters are as follows:
T1a T1b T2a T2b
0.6693245 0.3882727 2.1700352 1.9049669

######Changing ApproxVal for the EBTest function to a larger value as suggested in the EBSeq documentation does not change the outcome for any of these transcripts, for ApproxVal settings of 0.00001, 0.01, 0.1 or 0.5.