Evaluating removing batch effect tools with PCA or MDS?

Question

[Lluís R.]

It is the first time doing a meta-analysis on microarrys. I have two studies from two different xips I want to use. I used MergeMaid to combine them and now I am looking to correct the batch effect. I have done it with ComBat and removeBatchEffect. To asses the best batch removing tool I performe a PCA and a MDS plot. In both cases the PCA and the MDS show an improvement against not using it.

The PCAs are quite different, when calculated with ComBat and removeBatchEffect, due to the underlaying the different system. However the MDS are quite similar. Does this imply something about the batch effect?

PCA of raw data:

MDS of raw data:

PCA of combat correction:

MDS of combat correction:

PCA of removeBatchEffect:

MDS of removeBatchEffect:

[Ron]

I have worked with combat on RNA-seq data and I have definitely seen improvement in the overall results.the data can be compared after batch removing batch effects.

[Lluís R.]

Thanks Ron, do you know if it is better with MDS or PCA to asses the changes? I mean, both function ComBat and removeBatchEffect work find according to MDS but on a PCA only Combat performs better. And ComBat is quite more "aggressive" with data than other methods, that's why it is more effective in big batch effects.

[Ron]

I have not worked with MDS yet,so I cant comment on that.But,PCA gives good results.Whether be it library effects or just batches from different times,after doing combat.