Gene regulation network analysis mixed with other unrelated events

Question

[tujuchuanli]

Hi,

Here we investigated the regulation change of dysfunction of a TF (TF A) and try to construct the gene regulation network related to it in mouse model.

Our experimental design is to take two samples (two mouse as replicates habiting the dysfunction version of TF A, case sample) every two weeks for RNA-seq and the time spanning is 18 weeks (9 time points). Considering the relative long time spanning and the transcriptional changes due to unrelated events such as growing process and tissue development, we also have a control group which is normal mouse (habit the normal version of TF A, control sample) for RNA-seq and take sample for every two weeks as well.

My question is how to rule out the unrelated events and just see the consequence of dysfunction of my TF? My plan is to get fold change ratio (reads counts for one gene in case/reads counts for one gene in control) as indicator. So I get two gene fold change matrixes for each time point. However, my worry is that this kind of value is not suitable for normal network analysis such as WGCNA. Is it OK? Or do you have any suggestions?

Additionally, someone suggested me the following two ways:

1. Constructing two separate networks (case and control), then we can compare the topological difference between these two networks.

2. Combing all samples and constructing one network, then we can analyze and identify module and hub genes in case group.

How do you think about these two suggestions?

Thanks

Answer 1

[andres.firrincieli]

I am not going to comment whether or not a co-expression analysis is the "right" strategy, because I do not know anything about that TF and I never worked with mice.

However, here is what I can tell you about the analysis with WGCNA, in case you decide to go for it

However, my worry is that this kind of value is not suitable for normal network analysis such as WGCNA. Is it OK? Or do you have any suggestions?

A fold change does not tell you anything about how expressed a gene is within a specific condition. Therefore, for WGCNA I would use normalized read counts; VST-normalized read counts is what I usually use in WGCNA

Additionally, someone suggested me the following two ways:

1. Constructing two separate networks (case and control), then we can compare the topological difference between these two networks.
2. Combing all samples and constructing one network, then we can analyze and identify module and hub genes in case group.

Both strategies are completely fine, but if you are looking for defferentially co-expressed genes, then you should go for the option 1.

Answer 2

[tujuchuanli]

Thank you for your replying, just one further question about fold change. My original thoughts is that the effect of dysfunction of my TF are mixed with other unwanted events such as growing and tissue development along the time course. The fold change or ratio between the reads counts for one gene in case and in control could reflect the true effect of dysfunction of my TF, since ratio can get rid of the other effects (such as growing and tissue development along the time course) which could happen in both group (case and control). Is it not reasonable? BTW, do you have any suggestions that what can I do besides network analysis for this kinds of data?

Thanks.

Sorry for appending answers to the threads instead of adding comments, since there is nothing happened when I clicking the add comments button.

[andres.firrincieli]

TF are mixed with other unwanted events such as growing and tissue development along the time course. The fold change or ratio between the reads counts for one gene in case and in control could reflect the true effect of dysfunction of my TF, since ratio can get rid of the other effects (such as growing and tissue development along the time course) which could happen in both group (case and control).

That is out my area of expertise because I do not know the biology behind your system and I do not really understand what your are looking for.

If you really think that the use of "fold changes" is the way to go then, perhaps, a differential cluster analysis is the best choice.

Answer 3

[tujuchuanli]

Thanks a lot for your great help.

[Jean-Karim Heriche]

Please use the "add reply" button when replying to a comment. This keeps the discussion organized and avoids creating the impression that a question is answered when it is not.