Does it make sense to transform expression data to zscore after performing RMA?
1
0
Entering edit mode
24 months ago
jeni ▴ 60

Hi!

I am analyzing expression microarrays data and after reading CEL files I perform RMA in order to get background corrected, normalized and summarized data. I have seen that some people transform this corrected data to zscore in order to reduce noise between samples. So instead of performing the differencial expression analysis on the log2 expression values obtained after RMA, they perform the analysis on the zscore of these log2 values.

Could anybody explain if this processing makes sense?

Thanks!

microarray expression • 887 views
2
Entering edit mode
24 months ago

I see no reason to do this, particularly if the data is normalised via RMA and any problematic samples have been removed.

We may, however, transform the data to Z-scores prior to clustering.

Kevin

0
Entering edit mode

Thanks for your answer! Could you explain what do you mean by 'prior to clustering'?

0
Entering edit mode

For example, prior to hierarchical clustering and generation of a heatmap using the differentially expressed genes.

0
Entering edit mode

Ok, I see. And the purpose of that is a clearer visualization? I mean, the results of the differencial expression analysis or the clustering could be different if after performing the RMA I get the zscore, or the results should be the same but clearer to plot?

1
Entering edit mode

The Z-score transformation is primarily for visualisation, indeed, as everything is then centered around 0. Z-scores are also just more readily-interpretable to humans, as Z=1 is 1 standard deviation above the mean, 3 is 3 standard deviations, et cetera.

I actually never heard of anybody who performed Z-scaling after RMA for the purpose of the differential expression analysis itself. Perhaps you could share those papers?

0
Entering edit mode

Hello Sir,

My question is regarding log2(count+1) normalization and z-score normalization. I have used gene expression dataset of breast cancer from UCSC Xena repository. This dataset was log2 normalized already. However, in my preprocessing step, I applied z-score normalization as well. What would be the advantage/disadvantage of z-score normalization in this scenario?

Thank you.

Kevin Blighe Sir, I would be really grateful if you could shed some light on it, or point out some resource(s). Many thanks.