Is it possible to analyze differentially expressed genes without sample replicates using DESeq2?

Question

[Sarath Kumar]

Hi everyone,

I am trying to run mRNA seq pipeline (HISAT > FeatureCount > DESeq2). I have sample data without replicates (1 normal and 1 Tumor). Now I want to analyse differentially expressed genes between these two samples. I ran the pipeline(deseq2) and got results with some warning message. But I just want to know that is it the correct way to do or not?.

If not please let me know any other method to analyse Differentially expressed genes without replicates.

Thanks,

[ATpoint]

It is technically possible, but rather meaningless from a statistical standpoint. Please use google and the search function for further details. This question has already been asked numerous times.

[Martombo]

I thought this didn't work in DESeq2, but I saw they recently made it possible. Anyway, the whole point of a software like DESeq2 is to make the most of replicates data. If you don't have any, you'll lose a lot of information and results won't be very reliable.

Answer 1

[Friederike]

is it the correct way to do or not?.

No, it is not, hence the warning message. DESeq2, edgeR, limma et al. have fairly sophisticated ways of guessing what the most accurate expression value change of a single gene is. The methods all depend on replicates being present to provide information about the variability of the expression values between measurements of the same condition. If you have one sample per condition, there's no way to see potential variability. The fact that you still get results is based on DESeq2 just returning the bare minimum (and warning you about this).

Just imagine you did this analysis the old-fashioned way for your one single favorite gene of interest. No way would you believe the results that were based on one qPCR experiment in condition A vs. one qPCR experiment in condition B. Or maybe you would, in which case, you're fine. Just keep in mind that unless you have additional evidence to back those findings up, they may not hold up to repeated experiments.

[Sarath Kumar]

Thanks, Friederike. I would like to know what if we don't have replicates, Is there any other standard tool or methods available to analyse DEG?

[Friederike]

Not that I am aware of. The most commonly used way is probably DESeq, which you've done already. But it may depend on the question you want to address.

Answer 2

[Santosh Anand]

?DESeq gives you the answer

Experiments without replicates do not allow for estimation of the dispersion of counts around the expected value for each group, which is critical for differential expression analysis. The DESeq2 authors have decided to no longer support the data exploratory (with warning) behavior of previous versions, and so analysis of designs without replicates will be removed in the Oct 2018 release:DESeq2 v1.22.0, after which DESeq2 will give an error. The release DESeq2 v1.20.0 will give a deprecation warning, and then use a design with only an intercept to estimate dispersion, the behavior for all previous versions of DESeq2 and the DESeq package.

As it says, it was good only for exploratory analysis (with no statistical significance whatsoever). However, it was creating so much confusion with people using that as the final analysis that DESeq team has decided to discontinue it throwing an error in next Oct18 version.

Answer 3

[EagleEye]

If you like to use only logFC as cut-off (without replicates), have a look at GFOLD.

[Sarath Kumar]

Thanks, I'll look into it.