RNA seq expression of different genes in the same sample

Question

[riccardo]

Hello, I would like to know if with the RNA seq you can compare the expression of different genes in the same sample. For example:

SampleA

GeneA 10 GeneB 5

Can I say the GeneA is expressed two times more than GeneB with a good confidence? Or is there some bias when you do this kind of comparisons? Thank you.

Answer 1

[WouterDeCoster]

You will need normalisation, raw read counts don't correspond directly to the abundance of the transcript since longer genes will have more reads compared to smaller genes. But for the rest, you could probably do that if your read count is sufficient high. If a gene has only 1 count and another gene 3, then it's incorrect to assume the second has a three-fold higher expression, due to Poisson variability on your read counts.

[Morgan S.]

Hi WouterDeCoster,

I hope it is ok that I build onto your answer here, as I am also interested in comparing gene vs. gene in individual samples, not sample vs sample. I have 1 control sample (non-stressed) and 4 individual samples (stressed) and I want to compare gene A vs gene B in each of my individual samples. So I normalized my read counts with TMM while also supplying gene lengths to normalize to in NOISeq. For TMM I used my control as the reference myTMM_lengths <- tmm(assayData(myData)$exprs, long=(mylengths), lc=1, k=0, refColumn=1).

Then I just ended up plotting the normalized counts like this to visualize gene count differences.

My question is then since my data is normalized, can I say in sample 1 that gene A is 2x greater than gene B? Would a z-test be more meaningful? Is this analysis similar to what is done for sc-RNAseq?

Apologies if my question is elementary. I think I am confused because I am not wanting to do sample vs sample as most DE programs are set up to do as I do not have 2 or more "factors" to compare and I am trying to learn what meaningful analysis I can do with these normalized read counts.

Thanks in advance! Morgan

Answer 2

[huwenhuo]

This question can go back to the microarray era. At that time, there were way many factors can affect the readout signal for each gene from its real expression levels. First is the selection of probes, probe length, GC content, positions on the array et al. Second is the PCR amplification during the hybrid that may generate great bias. May have other concerns. In a word, this is not possible even to think about it.

The current sequencing techniques are definitely more quantitative. We have reads come out randomly from the whole molecular, we can normalize reads based on the gene length et al. Nevertheless, the concerns from the microarray era are actually most remain, to my personal view. I haven't found anyone to prove how quantitative the RNA-Seq is and whether they are good enough to compare the levels of gene a and b within the same samples. I would love to know this if anyone done such work.

Answer 3

[riccardo]

Thank you very much!

[WouterDeCoster]

Please use ADD COMMENT to answer to previous replies, as such this thread remains logically structured and easy to follow.