Test fold change significany using duplicate reference transcriptome (RNA-Seq)

Question

[songzewei]

Hi, All

Is it possible (or reasonable) to compare a single RNA-Seq sample (without replicate) to a duplicated reference transcriptome?

In our experiment, we have sequenced the control treatment in duplicate, but all the other treatment only once. I was wonder if it is possible to use the duplicate controls as a reference to infer the significance (p-value) of the fold change in other samples?

It seems to me it is a simple t-test, but I'm not sure if it makes sense. Also, is it possible for us to use DESeq(2)?

Thanks!

Zewei

Answer 1

[Matteo Schiavinato]

Hi Songzewei,

Is it possible (or reasonable) to compare a single RNA-Seq sample (without replicate) to a duplicated reference transcriptome?

First of all, you have to be aware that you won't have a strong statistical outcome out of this, but in principle it is possible if you state that you're not considering what comes out of the analysis as glittering gold. Basically, without replicates you cannot confirm values multiple times, and thus you will never know if that is a fluctuation of the data or a meaningful result.

However, there are some things that you could do: normalization for gene length and sequencing depth (FPKM or TPM) will at least make the numbers comparable between lanes. Also, if you run DESeq2 you might find some differentially expressed genes (i.e. having a log-2-fold change which is over or under your threshold - let's say +/- 1 or +/- 2). The only thing is that they won't have an adjusted p-value which allows you to say "ha! gotcha!".

To cut a long story short, you will reasonably find results, but you won't be able to tell if they're due to random fate or to an actual biological property of the sample (meaning of p-value).

Answer 2

[Ron]

Yes,its possible.

You can probably take the mean of control samples of all genes, and then calculate the log fold change as :

log(Test/Mean of controls )

You can get p -value as well by

pv.list <- apply(dat,1, function(x) tryCatch( 
  t.test(x[1],x[2],paired=TRUE)$p.value, error=function(x) NA ))

Where dat contains the FPKMS of Single sample in first column,and the mean FPKMS of control in the second column.

Ofcourse,its not an ideal scenario of not having duplicates of the test sample for calculating statistical significance but can give some meaningful results.

Answer 3

[Tonor]

Yes it is possible, but like everyone is saying you lose alot of power, and its important to remember how variable RNAseq is.

GFOLD is a good tool for comparisons without replicates: http://www.tongji.edu.cn/~zhanglab/GFOLD/index.html

You could compare your treated to each control separately and look for consistently up/down regulated genes, or combine you control reads into one and compare control vs treated in a single go.