How can I perform Differential expression analysis with just one control and one cancer sample?

Question

[k.kathirvel93]

Hi EveryOne,

Now i am in a situation to perform Differential gene expression analysis with just one control and one cancer sample. But, edgeR and DESeq2 will ask atleast two samples for control and cancer alone. What and How i can do now. Any suggestions. Thanks .

[GenoMax]

k.kathirvel93 : Please do not delete posts after they have received comments/answers.

[Joe]

How can you possibly analyse this data through DGE? They're telling you you need n > 1 for a reason.

The best you can do (to my mind), is map the reads, and just eye-ball the mapped reads to say "yes this gene expressed, and no this other gene didnt".

[russhh]

but even 'detection' isn't 'expression' so I wouldn't place any faith in an n of one for expression-vs-nonexpression. I'd refuse to do anything; but then I'm a massive grump. Tell your collaborators to analyse more samples.

And once you've done that, write up a code-of-conduct that states that any group / experimenter wishing to work with you can expect [all these benefits] but must adhere to [all these ethical and scientific standards] and must meet to discuss any planned experiment.

Expecting n > 2 in each arm is a prerequisite for publishing when using established experimental methods, so put that in your statement.

[Joe]

True, expression of the RNA is a long way from concluding a protein is doing something, but we have to suspend our disbelief somewhere :P

[russhh]

It wasn't so much protein I was referring to. To me the "ability to align something to a position in the transcriptome" is in no way the same thing as "reliably detecting the expression of the corresponding transcript". you are right though, sometimes you've just got to accept the limitations of your data and methods to get stuff off the desk.

[ATpoint]

You can do some rough data exploration like transforming the counts per gene to TPM, and then get the log2 fold change for every gene (cancer over control). This of course does not allow any statistics, does not inform you about the variation, and therefore is not reliable at all. Like every high-throughput NGS experiment, RNA-seq suffers from the meanvariance relationship, so genes with low counts are more prone to show high fold-changes. That is why replicates are so important. Keep that in mind.

Answer 1

[Damian Kao]

You can't calculate a p-value with no replicates as there is no measure of variance among your samples.

Best you can do is to filter out genes very stringently based on absolute read counts and then pick genes with high fold-changes.

Answer 2

[h.mon]

First, keep in mind all warnings posted here are very true: differential expression analysis without biological replicates is very poor experimental design.

Having said that, there are some software specifically designed to perform DGE analysis without biological replicates, like GFOLD and NoiSeq.

Finally, keep in mind all warnings posted here are very true: differential expression analysis without biological replicates is very poor experimental design.

Answer 3

[ewre]

You can't.

But you can try to calculate fold change and pick out genes of interest based on fold change. Be careful because it will be noisy.