# Negative scale-free topology

I am trying to build gene co-expression networks with WGCNA by combining in-house and publicly available RNA-seq datasets. I am interested in identifying gene networks associated with different environmental conditions. I used SVA to adjust for batch effects and "protected" different environmental conditions in the model as I don't want to lose information about them. However, when I used the "batch-adjusted" data to create networks, all the scale-free topology indices were negative (max = -0.05).

I tried generating the networks with softpower = 16 as it is suggested for sample sizes 20-30. However, only five networks were generated which is far less than I expected.

What might be the reason for getting negative scale-free toplogy indices?

Can combining studies from different environmental conditions cause the networks to be not biologically meaningful?

• Welcome to the site. I have some questions before being able to provide an answer: Could you show (the code) how did you modify your data to remove the batch effects with sva? How many samples and genes do you have? Are all the RNA-seqs built against the same genome version? How did you call the network construction, full automatic or manual? – llrs Jun 28 '18 at 21:40

Whenever you have batch or other nuisance variables driving variation you're going to get odd networks. You tried using SVA, which is a good method to use, but it seems in this case that it's able to optimally handle this dataset. I also strongly suggest that you use pickSoftThreshold() to choose a softpower, since that may produce better results (though since you already have negative values I'm not holding by breath). In general, you're going to need to play with either SVA or filtering to remove the batch-related genes that are causing problems.