I’m doing the meta-analysis on bunch of microarray data. There are from patients, with several time points before and after treatment. Actually, I’m interested in the post-treatment data (e.g. Day7), I wonder if reasonable only use normalized Day7 data to conduct the network construction, or in order to get rid of individual variances, should I process the data like Day7/Day0 and then use the processed data to do the WGCNA.

The other question is, it’s meta-analysis, I collect the data from serval different diseases but similar, from disease A, maybe have 8 dataset included in the analysis, but for another diseases, just have 1-4 datasets included. I thought, it may cause the bias to disease A (if I’m lucky to get the very conservative module), how can I consider the bias or include data set weight power to correct the bias?

The similar question is, each data set owns different sample sizes, should I and how to consider the weight of each data set?

If you can use the most homogeneous samples for WGCNA in order to find the accurate relationship between genes (specially if you have more than 12 samples in that group Day7).

If you are doing meta-analysis check the bias before proceeding any further. Check using PCA or similar techniques if the samples group by dataset and not by diseases. Normalize properly and take all the variables that are reasonable to normalize the data. Only then use WGCNA. In WGCNA you can't avoid this kind of bias but you could check the analysis with all the datasets and without dataset A, or balancing the number of samples for each disease/condition.