Hello,
this is my first question on the forum, so I apologize if i'm not doing things properly.
I'm a systems biology student currently working with WGCNA algorithm on a microarray dataset. I want to assess modules that are specific to one of my two conditions, by studying module preservation btw the two networks. My modules are defined based on the network of my condition of interest (and not the control one...don't know if it's correct to do so).
However, i'm quite struggling to find optimal parameters for module detection (which cutTree method, deepsplit, minModuleSize,...) so before running modulePreservation function btw the two networks, i wanted to validate the quality of modules.
To that extent, "Is My Network Module Preserved and Reproducible? Peter Langfelder, Rui Luo, Michael C. Oldham, Steve Horvath" mentions the following : "Module quality measures : Although not the focus of this work, we mention that a major application of density-based statistics is to measure module quality in the reference data (for example, to compare various module detection procedures)"
So, my firt question is : How to implement that, ie is there a way to say benchmark the parameters and find best combinaison using the criterions of modulePreservation function (separability, density..) I would like to know how is it possible to do so using the modulePreservation function in WGCNA. My second question is : the authors also state that this function can be used to check robustness of module definition while using it across the same data if i understood it right. How so?
I also have another question not directly related : is there a difference btw the two following approches, and if so, which one is more appropriate: 1)construct two separate networks and find less preserved modules to get insight into modules specific to the condition of interest 2) construct one combined network and then correlate modules to the conditions (recoded as dummy variable 0/1 since it is a binary trait here)
In addition 1)my tree looks quite strange and 2)leads to a huge grey module. Besides, observation n°2 seems to be the case when i construct a signed network(beta=18) while unsigned network(beta=8)
does not show such pattern. If someone could give a hint on the reasons why it could have such shape. ( the dendrogram clusters 10000 most varying gene / average hierarchical clustering / cutTreeDynamic).
I've mixed different questions here, i'm sorry. Thank you very much for your help.
Sara
Hi, for precision, I just wanted to cite another extract from the paper "Is My Network Module Preserved and Reproducible? Peter Langfelder, Rui Luo, Michael C. Oldham, Steve Horvath" in which the stategy I want to implement is explained. However I cannot manage to fully understand how to do the following : "A third use of module preservation statistics is to define measures of module quality (or robustness), which may inform the module definition. For example, to measure how robustly a module is defined in a given correlation network, one can use resampling techniques to create reference and test sets from the original data and evaluate module preservation across the resulting networks. Thus, any module preservation statistic naturally gives rise to a module quality statistic by applying it to repeated random splits (interpreted as reference and test set) of the data. By averaging the module preservation statistic across multiple random splits of the original data one arrives at a module quality statistic."
My ultimate aim would be to benchmark two parameters of module detection (deep split and merge Cut Height) and choose the best combination based on those module quality statistics. Is that possible ?
Thank you.