Bioinformatics Stack Exchange is a question and answer site for researchers, developers, students, teachers, and end users interested in bioinformatics. It only takes a minute to sign up.
Sign up to join this community
I have three bulk RNA-seq samples (test1, test2, test3) without replicates.
And I noticed that DEG analysis tools such as DESeq2/edgeR/etc cannot be applied for data with no replicates.
So, I just drew a heatmap with expression value (TPM) of genes of interest, using NMF (aheatmap).
But now I want to perform DEG analysis across the three samples. Is there anything I can try for the analysis?
My TPM data (.csv) looks like below.
Thank you so much.
According to DESeq2, you cannot analyze your data without having replicates.
In short, you have an experiment where you compare A to B because, in your theory, A is more expressed than B. However, you do the experiment only once. Without repeating the same experiment again under the same circumstances, there's no way to confirm you will have the same results and that your theory is correct.
You can still compare the ratio of A/B but that will not be meaningful in any way since there's not enough statistical power behind it. In your case, doing the ratio of test1/test2, test2/test3 and test1/test3.
Log2FC here since these values are not fold-changes. If you mean calculating the Log2 on the ratio of these values, you can do it, but it's only another way of representing your data. You have to be careful about the 0 since you cannot do Log2(0) -> undefined. Obviously, you also can't divide by 0 when performing the ratio. So you have to decide what to do with rows where there are 0. If this answer satisfies your question, don't forget to accept the answer please.
$\endgroup$
Sep 17, 2021 at 11:44
In this situation, you can apply the strategy in section 2.12 in the edgeR manual which basically "makes up" a dispersion estimate and then run the normal DE strategy. This is obviously neither reliable, nor publishable but at least it gives you a list of genes you can use for validation. Treat results with care, focus on genes with decent counts (not low on both conditions of the contrast) and select those with large effect sizes (fod changes). Alternatively, you can run vst from DESeq2 and then rank them by fold change, again focusing on genes with decent (non-low) counts. Both is not really reliable, so be careful with the results. As you have no information on the variability of the groups the fold changes are of limited value, but if you are forced to go with these data it is better than nothing.