Hi Zeynep,

First, thanks for your interest in this project. I used plate numbers in the GLM, and yes, batch effect can still be observed if you use all genes of the “normalized” dataset. I wasn’t aware that Synapse normalized for batch effect, but if so, I agree with you that this step did not successfully remove all batch effects.

The reason you don’t see batch effects in my second PCA plot is because of the feature selection step, where I used GLM to select a handfull of genes (~900) that are associated with case/control status while taking into account various confounders as covariates (including plate numbers).

Please let me know if you have further questions.