Using DESeq2 as one sample vs. population

Question

[hkarakurt]

Hello everyone, I could not find proper answer for my question online. I use DESeq2, limma and edgeR for standard control vs. disease comparison with multiple samples for each condition. Is it possible to test using only one sample of disease and compare it agains the all samples of control group? I asked for DESeq2 but I am open to all other possible options. I have count matrix but also can generate FPKM etc.

Thank you in advance.

[jv]

This topic is discussed widely online.

For a good starting point see: How to run the Deseq2 tool without replicates

[hkarakurt]

Thank you. As I understand is about analyses without any replicates. I have replicates for both conditions in fact but I am trying to do one sample test. I want to compare one disease sample against whole control samples. I will do the same for all disease samples and have p-values of genes for each disease sample.

[jv]

So you do have replicates of the disease condition? If that is the case, then can you elaborate on why it is that you want to do single sample differential expression analysis? Is there considerable heterogeneity across the disease samples? What question are you trying to answer by taking this approach?

If we can get a better understanding of what you are trying to achieve we may be able recommend alternative analyses that are better suited to the task.

[hkarakurt]

Yes actually I have replicates for both control and disease but my I am trying to apply a patient-specific differential expression analysis. In the end I want to have a p-value for each gene for each disease sample. Actually I am trying to do something like one-sample t-test but I do not want to use t-test here.

Answer 1

[ATpoint]

Technically yes, but be aware that the dispersion comes exclusively from the control samples. That means if the disease is very heterogeneous while the controls are not you will get many false positives and vice versa. Treat results with care, try to validate interesting hits.

[hkarakurt]

Thank you for your answer. How should I design my design matrix? Also which package do you think will be result with less false positives?

[ATpoint]

How should I design my design matrix?

If I understand correctly you have two groups, one with and one without replicates. That would come down to ~group.

Also which package do you think will be result with less false positives?

None, I guess. Lack of replication cannot be magically corrected. Getting experimental replicates for the unreplicated group is the only good option here I think. Else, get what DESeq2 returns and apply biological knowledge and validation. The point here is that with a single sample you simply do not know how heterogeneous this group is in reality, and I doubt there is any stats magic that will tell you this other than an actual experiment.

[hkarakurt]

Actually I have replicates for both of the conditions. Let's say I have 5 cancer and 5 control samples. But my aim is here to test each cancer sample against all control samples. Similar to one-sample t-test. In the end I want to have a p-value (or adjusted p-value) for each gene for each disease sample.