Hi Biostars community,
As a trait of interest, I use the categorical variable "trait" coded by 0/1 for Normal/Tumor tissue samples respectively. I have a matrix of expression data (Normalized and variance stabilized data) per gene (1900 genes) per sample (I've 22 Normal samples and 22 Tumor samples). I have also the trait data matrix with two column: sampleName and traitValue (0 for normal sample or 1 for tumor sample). I'm interesting to highlight modules in both Normal and Tumor status using WGCNA.
What I have already achieved:
- from the trait data matrix, I extracted a matrix of normal samples only (in this case, the traitValue=0 for all the 22 samples), and another matrix for the tumor samples only (in this case, the traitValue = 1 for all the 22 samples).
- from the expression data, I have formed two matrices of expression data (the first contains the normal samples, and the second the tumor samples)
- I have already detected modules in both sample groups.
The problem:
When I try to identify the correlation (and significance p_value) between the modules and the trait within each sample group using the following command:
moduleTraitCorrelation = cor(MEs, datTraits, use = "p");
since the datTraits variable represents a vector of equal values (0:normal, 1:tumor) so absence of variance, I get NA values in the resulted correlation matrices (one column represents module names, and another the correlation values)
any idea ?
You can compare the module preservation in both datasets to see if the co-expression is kept. Or you could build new modules with all the dataset and then correlate with the traits (this way you will have variance on the trait)
Im curious " I use the categorical variable "trait" coded by 0/1 for Normal/Tumor tissue samples " is the design correct because let's say we have to define cancer in terms of stage ,it can be of varying degree so if i give Tumor as 1 it will be 1 irrespective of what stage it is ? would be glad if you can help this out since im facing the same problem with my trait design as no patient data is available but i know which are normal and tumor samples
The way that it is described in the tutorials of WGCNA, yes tumor 1 otherwise 0, but you could also have another variable that is stage, and that would be 0 no tumor, 1 small, 2 bigger, 3 metastasis or whatever.
okay sounds reasonable ,since i have no patient data ,but the data im using it tells about the marker they have used to purify those population, so in my trait data what im doing is i have taken the raw counts of all those maker and made them as my trait.so is that the right approach to address the problem , because what i get is based on cell surface marker you can define cell .so let me know if im going right way or not...
If you are going right or not might be better addressed by your supervisor. But it seems reasonable to me
thank you for giving me confidence let me see if i get any interesting result
Did you solve your problem? Because I got the same issue with NA values when I check the correlation between my traits and MEs. Since all my traits are 1.