What statistical analysis to perform on data from gene expression analysis?

Question

[DanielC]

Dear All,

I performed hierarchical (HCL) and Self-organized map (SOM) clustering on some cancer datasets, to identify the co-expression pattern of the genes involved in the cancer. For instance, I got data like this:

There are 25 datasets from GEO database (like GDSXXX) of microarray mRNA expression values. In each dataset, I looked for the co-expression pattern of a set of gene-pairs using HCL, and got results like this

pair1: gene x<--> gene y (found co-expressed) in all datasets from dataset 1 to dataset 25
Pair2: gene r <--> gene z (found co-expressed) in just 2 datasets among all datasets from dataset 1 to dataset 25
.
.
.
.
Pair 16: gene u <--> gene o (found co-expressed) in all datasets from dataset 1 to dataset 25

Similarly, I performed the co-expression analysis using SOM and got results like this:

pair1: gene x<--> gene y (found co-expressed) in all datasets from dataset 1 to dataset 25
Pair2: gene r <--> gene z (found co-expressed) in just 5 datasets among all datasets from dataset 1 to dataset 25
.
.
.
.
Pair 16: gene u <--> gene o (found co-expressed) in 10 datasets from dataset 1 to dataset 25

Can you please let me know what statistical analysis can I apply here to find out the most significant co-expressed gene-pair(s) in both HCL and SOM? Thanks much,

DK

[Devon Ryan]

Are you doing this all manually or as part of the WGCNA package?

[DanielC]

Thanks for the reply! I used R to perform the HCL and SOM clustering. I have not used WCGNA package before. At this stage, am looking for what statistical analysis to perform to find out the most significant co-expressed gene-pair(s) in both HCL and SOM generated data as mentioned in above example. Could you let me know what analyses could be done for such aim? Thanks.

[Devon Ryan]

The WGCNA method to do that would be to look at the degree of module membership, or perhaps correlation between the genes and the module eigengene.

[DanielC]

Thanks! Could you please let me know what you mean by "module" here? or if you could guide me to an appropriate source. Thanks!

[Devon Ryan]

You've essentially created a network graph based on coexpression/covariation. One can subdivide that graph into covarying sections ("modules") that can then correlate to various things such as phenotypes and treatments.