DESeq2 merged vs unmerged replicates

[spabinger]

Hi,

I am currently analyzing a RNASeq experiments with 2 conditions. Each condition has several biological replicates and each replicate was sequenced multiple times (on different lanes and even flowcells). For each sub-replicate I have a separate fastq file.

Control 1-1 Treatment 1-1
Control 1-2 Treatment 1-2
Control 1-3 Treatment 1-3
Control 2-1 Treatment 2-1
Control 2-2 Treatment 2-2
Control 2-3 Treatment 2-3
...

I performed now 2 analyses:
1) Fastq -> STAR -> htseq count -> DESeq2
2) Fastq -> Merge Fastqs of each replicate (1-1+1-2+1-3) -> STAR -> htseq count -> DESeq2

I checked the count files and the individual counts of the replicates sum up to the merged count file.

However, when I run now DESeq2 on the different count-data-sets I get different results (most notably the adj p-value).

Not merged:

                    baseMean log2FoldChange      lfcSE      stat        pvalue         padj
                   <numeric>      <numeric>  <numeric> <numeric>     <numeric>    <numeric>
ENSG00000143369.10 479.26267      6.1618396 0.28753411  21.42994 7.026598e-102 2.636520e-97
ENSG00000135250.12 381.34094      0.7005228 0.03782914  18.51807  1.476331e-76 2.769745e-72
ENSG00000163898.5   62.74118      5.4375752 0.30318159  17.93504  6.281376e-72 7.856327e-68
ENSG00000154269.10  65.19322      3.9433545 0.23587129  16.71825  9.651700e-63 9.053777e-59
ENSG00000108821.9  306.78844      3.0304242 0.18264854  16.59156  8.020919e-62 6.019218e-58

Merged:

                     baseMean log2FoldChange     lfcSE      stat       pvalue         padj
                    <numeric>      <numeric> <numeric> <numeric>    <numeric>    <numeric>
ENSG00000077327.11   63.88176     -3.1465321 0.5082833 -6.190509 5.997039e-10 1.553113e-05
ENSG00000222022.1    13.70911      3.2297638 0.5455883  5.919782 3.223688e-09 4.174354e-05
ENSG00000135250.12 3787.92187      0.7295915 0.1324829  5.507063 3.648689e-08 3.149791e-04
ENSG00000108821.9  2925.22227      2.4831871 0.4592653  5.406869 6.413598e-08 4.152484e-04
ENSG00000110427.10   74.67557      2.7264235 0.5144769  5.299409 1.161784e-07 6.017577e-04

Could you please tell me if that is expected?
Which approach should I use?

Thanks for you help,
Stephan

[dpryan]

The difference is because DESeq2 thinks your technical replicates are biological replicates. If you really want to include them then the best you can do is to make a factor for the flow cell. I have yet to see an instance in the last few years where there's a lane batch effect (other than getting fewer reads...but that's not a batch effect). In reality, you might just do a PCA plot and if there's no real batch effect then go with the merged results.

[spabinger]

Thanks for the explanation.

In my case, I'll go for the merged files, correct?

Thanks,
Stephan

[dpryan]

Correct

Ignore me: this message is too short otherwise.