How to calculate if statistically a variable of a bulk RNA-seq affects the comparison of interest?

Question

[Rafael Soler]

Hello,

If I want to make an RNA-seq comparison of WT vs KO, is there a way to statistically check that the gender variable (FEMALE vs MALE) does not affect the main comparison (WT vs KO)? In the PCA we can clearly see that PC1 is the comparison between SP vs CX, and PC2 is WT vs KO, but there are no big differences in relation to gender, would this be sufficient justification for not taking it into account when modeling? Or should a statistical test be performed?

Best,

Rafael

[Shred]

Already answered here Adjusting for confounding variables in RNA-SEQ Differential Expression Analysis

[Rafael Soler]

Thanks, I couldn't find it.

Is there a way to perform this logistic regression but instead of a single gene, with all genes? The intention of the analysis would be to see if sex has a global effect between WT vs KO, not on how many genes it does and how many it doesn't Kevin Blighe

[Rafael Soler]

Like this?

MyData                      
Schizophrenia   Gene1   Gene2   Gene3   GeneX   Age Sex
Y   2.33    3.33    1.33    …   45  M
Y   3.21    1.11    2.21    …   43  M
N   1.21    0.2 3.31    …   26  F
N   2.11    2.21    4.41    …   35  T
...     

[Kevin Blighe]

The main difference is along PC1 for CX and SP. The percent variation tells me that you need to adjust for this somehow, via one or more of these options:

1. stratify dataset by CX and SP, i.e., run separately for each
2. adjust for CP|SP when deriving test stats. for KO vs WT, i.e., treat CP|SP as a covariate / confounder
3. use an interaction model formula

I see no justification based on the PCA bi-plot to do any adjustment for sex.

[Rafael Soler]

Thank you Kevin!

Although in the PCA it is clear that the only thing that should be avoided is the effect of CX/SP, could it be statistically analyzed if the "Sex" variable has an effect on the WT vs. KO comparison?

Best,

Rafa

[swbarnes2]

With only 4 samples, its hard to separate out the contribution caused by individual variation versus variation caused only be sex.

[Rafael Soler]

Thanks. Because it has a n=2?