Module- Triat relationship in WGCNA

Question

[Prakash]

Hello all,

I am working with 20 samples of RNAseq data ( 5 control and 5 KD with five different stimulation) and have identified several module using WGCNA. Next I would like to perform module- condition relationship analysis, basically I would like to identify condition specific module. I have designed the condition in binary format (see table below). can someone tell that the design is correct or not, if not, what would be best way to design trait in this scenario.

sample  EU  EI  EC  EP  EA  NU  NI  NC  NP  NA
E_Uns_R1    1   0   0   0   0   0   0   0   0   0
E_Uns_R2    1   0   0   0   0   0   0   0   0   0
E_IFNg_R1   0   1   0   0   0   0   0   0   0   0
E_IFNg_R2   0   1   0   0   0   0   0   0   0   0
E_CpG_R1    0   0   1   0   0   0   0   0   0   0
E_CpG_R2    0   0   1   0   0   0   0   0   0   0
E_pIC_R1    0   0   0   1   0   0   0   0   0   0
E_pIC_R2    0   0   0   1   0   0   0   0   0   0
E_all3_R1   0   0   0   0   1   0   0   0   0   0
E_all3_R2   0   0   0   0   1   0   0   0   0   0
N_Uns_R1    0   0   0   0   0   1   0   0   0   0
N_Uns_R2    0   0   0   0   0   1   0   0   0   0
N_IFNg_R1   0   0   0   0   0   0   1   0   0   0
N_IFNg_R2   0   0   0   0   0   0   1   0   0   0
N_CpG_R1    0   0   0   0   0   0   0   1   0   0
N_CpG_R2    0   0   0   0   0   0   0   1   0   0
N_pIC_R1    0   0   0   0   0   0   0   0   1   0
N_pIC_R2    0   0   0   0   0   0   0   0   1   0
N_all3_R1   0   0   0   0   0   0   0   0   0   1
N_all3_R2   0   0   0   0   0   0   0   0   0   1

Thanks

Answer 1

[Lluís R.]

Your conditions seems like blocking factor. In the WGCNA it is recommended to make the correlation between the eigenvalue of the module with the phenotype variables. I would recommend however to do the correlation between the eigenvalue of the module with the eigenvalue of your condition.

However, if this is are really 5 different conditions it might be worthless, as the co-expressed genes might be few or not highly co-expressed...

[1769mkc]

but in this paper they have take three condiion in the last part of their analysis and did WGCNA can you explain ,how did they perform ?

[Lluís R.]

Well I don't know anything from this paper beyond what it is described, but looking for the Figure 4 a. it seems like they made the correlation between the condition and the eigengene. If it is accurate or not is another thing... You can do as them, but note that they have more than 4 samples for each condition and up to 9.

Upon closer look to your matrix, it seems like you have several factors happening at the same time. If I understood correctly you have 2 conditions (N and E) and then (Uns, IFNg, pIC, CpG and all3) with two replicates for the second condition, which results in smaller groups for each condition, which are too small to do something with confidence.

[1769mkc]

yes the sample size is a concern ..