Hi, I am using DEXSeq for testing DEU. The DEXSeqDataSet is converted from SummarizedExpriment instance of containing ranges and counts for exons. The covariates used in the design model is Treatment representing two conditions (DOX and NODOX), each of which has three triplicates.  I notice there are some exons expected to have differentially usage. The counts in one condition is relatively larger than the other. But the pvalue is almost one. I do not understand why, and perhaps you guys can give me some insights and suggestions? 

Below is chunk of my data:

> dxr[name,]

LRT p-value: full vs reduced

DataFrame with 3 rows and 13 columns

                 groupID   featureID exonBaseMean  dispersion         stat

             <character> <character>    <numeric>   <numeric>    <numeric>

7863:E292716        7863     E292716    69.867123 0.001561678 -0.054938652

5789:E214197        5789     E214197     4.986662 0.316687994  0.008410802

8215:E303543        8215     E303543   161.418531 0.000916123 -0.003169622

                pvalue      padj     NODOX       DOX log2fold_DOX_NODOX

             <numeric> <numeric> <numeric> <numeric>          <numeric>

7863:E292716 1.0000000         1 0.6176518  2.652094           2.102266

5789:E214197 0.9269281         1 0.5525945  1.190415           1.107172

8215:E303543 1.0000000         1 0.3627619  4.178818           3.526000

                             genomicData       countData transcripts

                               <GRanges>        <matrix>      <list>

7863:E292716 chr3:+:[93820935, 93821042] 114 134 155 ...    ########

5789:E214197 chr1:+:[30616742, 30616852]      7 2 20 ...    ########

8215:E303543 chr3:-:[94178072, 94178182] 315 288 324 ...    ########

The counts for NODOX (column 4-6) are almost zero whereas DOX (1-3) have more counts:

> counts(dxd)[name, 1:6]

             [,1] [,2] [,3] [,4] [,5] [,6]

7863:E292716  114  134  155    0    0    0

5789:E214197    7    2   20    0    0    0

8215:E303543  315  288  324    0    0    0

The size of lib is sufficient:

> colSums(counts(dxd)[, 1:6])

[1] 19786058 17779995 18319056 15141153 22307765 20236315

Is there anyway I can fix the problem with modifying any parameters of DEXSeq? 

Thanks,

Chao-Jen

Hi Chao-Jen Wong,

Thanks for adding the plots. Your exons have indeed a huge difference in counts between the different conditions. However the difference seems to be related to differences in gene expression rather than differences in exon usage (see the DEXSeq vignette and paper for details). This is the reason why they don't appear to be significant.

Alejandro