Hallo everyone!

I have encountered a problem while working with Agilent 4x44k whole mouse genome v2 microarray chips.
A part of my experiment was performed 1.5 years ago (pilot study), and the second part - just now.

All the data obtained from the both parts of the experiment has been background-corrected and normalized with limma. I performed a backgroundCorrect(method="normexp") and than used normalizeBetweenArrays(method="quantile").

The data from different parts of the experiment forms two distinctive clusters ("new" and "old") in the Principle component analysis (PCA), even though the samples are supposed to be the same and come from the same origin.

Did i oversee some requirements? Are the normalization/background correction methods more suitable to this kind of problems? Thank you!

You won't be able to normalize out a batch effect, and it is not unexpected to see the two batches cluster like that. If  you ran all the groups you are planning to compare in both the pilot and the second batch, then you can a.) fit a batch effect in your linear model or b.) use ComBat() in the sva package to remove the batch effect. There are examples of doing both things in the users guide for limma, and in the sva vignette, plus a search for batch effect on this support site will turn up even more information.

If you ran one or more of your samples in just one of the batches, then you have unwittingly confounded biological differences with your batch effect, and there is no way to fix that.

Regarding:a.) fit a batch effect in your linear model or b.) use ComBat() in the sva package to remove the batch effect.  

After removing batch effects using linear model the output data will be the residuals, what is the output form of the data when using ComBat? Residuals? Can you clarify, please?

Thanks!