I would like to know if it will be correct to perform normalisation of raw microarray data on every pair of samples to be compared (eg, control vs treatment 1, control vs treatment 2, etc) and then use the data to identify differentially expressed genes. I think normalisation should be performed taking all samples (controls, treatment 1, treatment 2, etc) together, and then use the data for identifying differentially expressed genes.
I would agree with Erik, but remember what assumptions are underlying this approach. You are assuming that approximately the same number of genes are 'on' and 'off' in all samples. If you have a strong reason to doubt that, and to think that one of your contrasts has generally higher expression in both treatment and control than another contrast, for some reason, then you might think about normalising them separately. You could check this out by normalising them all together first, and looking for evidence that your assumptions were safe.
I would perform normalization of all samples together, and then do the comparisons to identify differentially expressed genes (DEGs)... otherwise, comparing DEGs lists obtained by the various comparisons would not be correct...
I would agree with Erik, but remember what assumptions are underlying this approach. You are assuming that approximately the same number of genes are 'on' and 'off' in all samples. If you have a strong reason to doubt that, and to think that one of your contrasts has generally higher expression in both treatment and control than another contrast, for some reason, then you might think about normalising them separately. You could check this out by normalising them all together first, and looking for evidence that your assumptions were safe.
My samples are related as it consists of control (uninduced) cells, induced cells, and post-induced cells (i.e. inducer removed and cells cultured for additional days without inducer). So my approach of normalising them all together is correct.
Thanks Erik, Jonathan and Rohan.
Forget normalisation and perform PCA having as variables the samples and as statistical units the genes, if you are working on the same tissue 'geenral size' effects will go with PC1 (all positive loadings) ans you will discover statistically relevant effects (if any) from the loadings of PC2 onward:
http://www.biomedcentral.com/1471-2105/7/194
http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0013668
- Badges
- Science topic
- Similar topics
- Bioinformatics
- Bioinformatics and Computational Biology