RNA-seq normalization in samples with different cell densities

[rpeiro]

Hi all,

We are going to make an experiment involving D melanogaster samples each of wich include about a 25 % of cells being part of a differentially expressed tissue. We are afraid that changes in genes with low expression levels can be masked due to this low representation of 25 %.
Does anybody know what is the more accurate normalization method?
Thank you in advance

[jparsons]

I have been working on a method to "unmix" RNA-seq data from mixtures for awhile, and I think it may be of some help to your situation.

You would need to:
*Be able to collect a sample of your "unwanted" tissue which was free of contaminants
*Be able to measure the mass of ribodepleted RNA (this can be done as a part of the sequencing experiment using spike-in RNA or by using more traditional means)
*Be willing to accept that your uncertainty "about a 25%" will propagate

If these options are possible for you, we can discuss details.

[rpeiro]

Thank you for answer me jparsons. Regarding with the points you posted:

*Be able to collect a sample of your "unwanted" tissue which was free of contaminants

We planned to sequence WT tissue as a control

*Be able to measure the mass of ribodepleted RNA (this can be done as a part of the sequencing experiment using spike-in RNA or by using more traditional means)

The selection of RNA to sequence will be using polyA tails

*Be willing to accept that your uncertainty "about a 25%" will propagate

An amount of uncertainty is inherent in this kind of experiments. Lets hope we be able to decrease it.

Thank you

[jparsons]

polyA selection is still filtering total RNA into mRNA and must be accounted for.

It still sounds like you may have an ideal experiment for a mixture design.

Call your WT tissue A, your modified tissue B, and your experimental samples C.

Your expression values in C will be: C = (A*.75*mrnaA)+(B*.25*mrnaB)

You could directly measure the mRNA content qRT-PCR, or determine it using spiked-in RNA. We calculate it as sample reads per microgram of total RNA divided by spike-in reads per microgram of spike-in RNA. This calculation emphasizes that the mRNA fraction is a correction for the differential enrichment between polyadenylated spike-in RNA and total RNA, which is only partly composed of mRNA.

The ratio of spike reads/sample reads in your mixture sample C is related to your spike/sample reads in samples A+B in the same way, so you can back-calculate the mrnaB from the mrnaC.

Again, the main limitation here is how well you are able to estimate the proportion of the tissues in your mixture samples. If you have more than 5% error in your proportion estimate(which is where the .75 and .25 come from), this analysis may be worse than the alternative.