Transcriptome analysis principle

Question

[2012201024]

Hi all,

In my opinion, I would like to consider the transcriptome as multiple QPCR. But in the actural transcriptome analysis process, there are no ck genes, like actin et al. So I want to know how can we confirm the different treatments have a same comparison background?

Many thanks in advance.

Answer 1

[Michael]

Please improve your post, I could hardly understand what you are referring to. I guess you are referring to "Normalization"? All transcriptomic techniques, be it microarrays, qPCR or RNA-seq, need some kind of normalization, you should therefore possibly read some review papers on RNA-seq normalization, transformation and differential expression. There are some good ones in this collection: http://www.mendeley.com/groups/617481/rna-seq/

In detail, my criticism of your way to approach this topic:

In my opinion, I would like to consider the transcriptome (sic, guess its RNA-seq) as multiple QPCR.

That's a very bad model for various reasons. (Can you see the differences?)

But in the actural transcriptome analysis process, there are no ck genes, like actin et al.

Do you mean house-keeping genes?

So I want to know how can we confirm the different treatments have a same comparison background?

I couldn't grasp what you mean by this, first it is better to talk about samples (not everything is a treatment), what do you mean by comparison background?

Answer 2

[Damian Kao]

You are asking how RNA-seq data is normalized? If there isn't a house-keeping gene as a standard, like in qPCR, how do you compare various samples?

The simple answer is that with RNA-seq, we normalize our libraries by making an assumption that the majority of our genes are not differentially expressed. Given that assumption, some measure of centrality in our expression distribution can then be used as the "standard".

For example, DESeq normalizes by first constructing a comparison dataset which is just the geometric means of each gene from each sample. Then for each dataset, the median of the ratios of each gene to the comparison dataset is taken as the normalization factor. In essence, it's a way to find the centrality of the dataset. The portion of the dataset that probably isn't differentially expressed.

EdgeR's TMM method uses multiple methods to filter out extremes of the dataaset (for example the bottom 5% lowest expressed genes, top 5% highest expressed genes) and then take the mean of the left over as the standard.