how to exclude/include microarray chip based on pseudo-images?

Question

[Juan Cordero]

Dear Biostars community,

I have got some affy microarray data to analyse, and some chips display artifacts, according to their residual pseudo-images. However, I do not know to what extent I should consider an array as acceptable based on the (size, shape...) artifact.

I have seen there are some methods for correction of such artifacts, either at probe level or at normalisation step, but I need a piece of advise to know If I should try to correct the chips or directly to discard them.

Thanks

[Jean-Karim Heriche]

I don't think there are any hard rules for judging acceptability of an array based on its pseudo-image. If there's sign of extensive artifact or systematic pattern then the array should be discarded.
In general, if I can afford it, I prefer discarding problematic data over trying to find a mathematical/computational fix. If possible, it's better to redo an experiment than working with dodgy data.

[Juan Cordero]

My problem to which extent of shape and size you would include an array in the analysis. For example, in this picture, there is a "drop" next to the border. In my opinion, it's too big, but when normalising the data, this array seems to be normal. However, there is another array, with a smaller "drop", which appears a bit abnormal, according to intensities distribution and RLE/NUSE. Then, the question is: if I discard the array with the small drop according to QC metrics, shouldn't I discard the one I show here, although it seems correct to the eyes of QC metrics?

Thanks