Always evaluate non-experimental chip effects correction (batch effects)??

Question

[Leite]

Hello all,

I wonder if always when alanalizing microarray data (affymetrix microarray) have been evaluated non-experimental chip effects corrections, also known as batches effects corrections. Batches can be in the form of different chips used for hybridizations, different days of biotin labeling etc.

In different works I've read that used affymetrix microarray I didn't see in the analysis codes any part related to Batch correction.

I thought it was necessary when using different GEO dataset (even if same chip) or different chips.

What do you say about that? Can not correcting a dataset downloaded by GEO generate unreliable data?

Best

[Benn]

The advantage of a batch correction is that you are sure that the results are what you test (treatment versus control) instead of differences of other factors such as between batches (which are not biologically relevant or interesting). So the results would be more accurate. Even if the batch effects are negligible, it doesn't harm to include them in the model.

[Leite]

Dear @Benn

Thanks for your answer,

So the data got more accurate, but not making the correction, wouldn't it be wrong?

If you finished an article and at the end noticed that the data did not have batches effects corrections. Even with the results agreeing with what you understand of the disease and what the literature, would I redo all microarray analysis and all the work?

[Benn]

If an article gets published, it doesn't necessarily mean they have used the best methods. If possible you can redo the analysis (if GEO provided enough info about batches, etc.). Nothing wrong with being critical, and knowing what you are doing.

[Kevin Blighe]

I have been re-processing a few dozen GEO microarray datasets in the past few months and notice quite 'severe' batch effects in more of these than I would expect. It is unclear to me if these effects were taken into account during the initial study.