Hi everyone,
So I ran the same transcriptomic data in the end of last month and today and, somehow, the differential gene expression results are different. I've checked the R version, the package versions and everything matches. The only thing I changed in my PC was an update of the Java version. Is this an expectable phenomenon? Or should the problem be coming from a different version?
Cheers,
André
Java version wouldn't change the results. Can you reproduce different results with identical R, Bioc and DESeq2 versions? Likely there was a change in DESeq2 version that may have gone un-noted.
Hi Michael,
I'm sending you the SessionInfo of both iterations, I didn't see anything different.
March SessionInfo():
R version 3.5.3 (2019-03-11)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 7 x64 (build 7601) Service Pack 1
attached base packages:
[1] parallel stats4 stats graphics grDevices utils
[7] datasets methods base
other attached packages:
[1] gridExtra_2.3 vsn_3.50.0
[3] stringr_1.4.0 xlsx_0.6.1
[5] genefilter_1.64.0 ggplot2_3.2.1
[7] org.Mm.eg.db_3.7.0 AnnotationDbi_1.44.0
[9] RColorBrewer_1.1-2 pheatmap_1.0.12
[11] DESeq2_1.22.2 SummarizedExperiment_1.12.0
[13] DelayedArray_0.8.0 BiocParallel_1.16.6
[15] matrixStats_0.54.0 Biobase_2.42.0
[17] GenomicRanges_1.34.0 GenomeInfoDb_1.18.2
[19] IRanges_2.16.0 S4Vectors_0.20.1
[21] BiocGenerics_0.28.0
April SessionInfo():
R version 3.5.3 (2019-03-11)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 7 x64 (build 7601) Service Pack 1
attached base packages:
[1] parallel stats4 stats graphics grDevices utils
[7] datasets methods base
other attached packages:
[1] gridExtra_2.3 vsn_3.50.0
[3] stringr_1.4.0 xlsx_0.6.1
[5] genefilter_1.64.0 ggplot2_3.2.1
[7] org.Mm.eg.db_3.7.0 AnnotationDbi_1.44.0
[9] RColorBrewer_1.1-2 pheatmap_1.0.12
[11] DESeq2_1.22.2 SummarizedExperiment_1.12.0
[13] DelayedArray_0.8.0 BiocParallel_1.16.6
[15] matrixStats_0.54.0 Biobase_2.42.0
[17] GenomicRanges_1.34.0 GenomeInfoDb_1.18.2
[19] IRanges_2.16.0 S4Vectors_0.20.1
[21] BiocGenerics_0.28.0
Regarding the "loaded via a namespace (and not attached)" packages, that I didn't add, there are indeed some differences. April's run has this packages, that are not present in the March one:
ashr_2.2-32
codetools_0.2-16
doParallel_1.0.14
foreach_1.4.4
iterators_1.0.10
MASS_7.3-51.3
mixsqp_0.1-97
pscl_1.5.2
SQUAREM_2017.10-1
truncnorm_1.0-8
Do you think this may have an effect?
Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
version 2.3.6
I just did a more thorough analysis on the results and I found the following:
There are differences in p-values and log2FC in the results obtained from DEG analysis from one DESeq model ran in March and the same model ran in April. This happens with the same R, Bioc and DESeq2 versions installed;
However, when the DEG analysis from the March Model is ran now, the results now match.
Looking into the 2 models, the counts normalized and raw, as well as SizeFactors, are the same. However, the values for dispersion are different; by subtracting the dispersion of the April Model from the March one, we get the following results (statistics summary):
I expect these differences are responsible for the incongruence of the results. Is there any reason for this to happen?
I'd be surprised if you really got even numerical differences from the same code base. But at this point, if you say that your March results must be tied to your March sessionInfo(), then I don't have any more ideas.