Valid comparison of gene expression between several genes in several cell lines

Question

We have gene expression data (Affimetrix mRNA gene expression results) for several cell lines, over a set of genes. Our goal would be to be able to compare relative gene expression for genes over the different cell lines.

The method we have thought of so far involves calculating some sort of gene-based average (the expression of a particular gene over all the lines) and normalizing our expression data, grouped per gene, based on that. However, we are not sure that this is a valid method for obtaining relative expression data from our data set, especially so that we can compare the data from several lines.

Does this seem like a valid strategy? How is this normalization commonly done? Should we do something differently to get a better comparison?

Thanks so much!

Answer 1

This is not an easy task, as there are many factors which you have to take in account. First there are intra-essay differences as slightly different conditions for each set of probes for the repetitions. Then, there are inter-assay differences, as differences in the array slides and so on, this paper ("Analysis of microarray gene expression data") goes into more details.

This articles might be helpful as well:

• Multiple-laboratory comparison of microarray platforms
• Increasing consistency of disease biomarker prediction across datasets.
• The curse of normalization.
• Microarray probe expression measures, data normalization and statistical validation.

What you could also do is go over the Researchgate, I think there are more people around with a lot of experience in this field.

Answer 2

If your data came from a single lab and a standard protocol gcrma normalisation is a good method to use (and comes with the affy R package); if the arrays are of U133A or HGU133plus2 types frozen RMA may be a good method to use. Finally you can check for and control for batch effects using ComBat and use the corrected expression data.