What lfcSE is dependent upon?
1
0
Entering edit mode
Sam ▴ 10
@sam-21502
Last seen 5 weeks ago
Jerusalem

I am looking at an experiment with 5 comparisons (and not much difference among the conditions). I tried separating out one comparison from all the system, running DESeq2 on it separately. This resulted in lowering the lfcSE for some genes, making more genes significant.

1) Are there any indications for discerning whether these additional significant genes are real or false positives?

2) I looked at the normalized counts for the analysis which includes all conditions, as well as the one where there is only one comparison. The expression values are very similar. What can explain the difference in lfcSE?

The analysis with the compared two conditions (treatment & control) only

Control1     Control2   Control3    Control4    Treat1      Treat2
71461.7      81214.1    89435.5     69952.7     122438.9    115726.5

baseMean    log2FoldChange      lfcSE               stat                pvalue          padj
91704.9     0.622783439312684   0.103093524641065   6.04095593279012    1.532038E-09    1.01581793E-05


For the analysis with all conditions ( there are other conditions besides treat & control, but I omit them here):

Control1    Control2    Control3    Control4    Treat1         Treat2
65019.9     74355.1     85325.1     67412.9     112764.2       110093.7

baseMean            log2FoldChange      lfcSE               stat                pvalue              padj
84231.6928842715    0.621226842710026   0.313839691139753   1.97944001427593    0.047766487091329   0.998357130543871


The values of the samples are not very different, yet there is a large difference in the stat and pvalue (based on a difference in lfcSE). What could be the cause for that?

3) What is the lfcSE dependent upon, generally speaking? I admit that the 2014 is above my level in statistics.

DESeq2 • 153 views
1
Entering edit mode
@mikelove
Last seen 1 hour ago
United States

I tried separating out one comparison from all the system, running DESeq2 on it separately.

This question is a FAQ in the vignette, take a look.

What can explain the difference in lfcSE?

The SE formula is in the 2014 paper in Methods section and it's worth examining if you are actually interested in what parameters influence it.

The SE depends on the gene's dispersion estimate which is not the same across these models, again see the FAQ on this in the vignette.