Important genes are not differentially expressed

Question

[bionewbie]

I am working on an experiment where the Genes of interest were not significantly expressed after the DESeq2 analysis. I applied the filter of 0.05 significance p-value and the genes did not appear. The important genes had a p-value of 1. Is there anything else that could be done as a Bioinformatician to proceed with the analysis?

Edit: My experiment design consists of a control and 2 mutation types. I tried to create DESeq2 conditions MA vs WT and MD vs WT. I am getting upregulated and downregulated genes but it does not consist of the genes that I am interested in. The genes that I am interested in had a log fold change value between 0 and -1 and an adjusted p-value of 1

[rpolicastro]

Assuming your code was correct, there isn't much you can do.

[Santosh Anand]

The only thing you can do is to look back at your hypothesis - as your data strongly suggest that your hypothesis is not correct!

Answer 1

[jared.andrews07]

p-value "hacking" is discouraged and can leave you chasing ghosts when you try to experimentally validate or follow-up on your findings. Without knowing anything about the experimental design, the only real advice I can give is that there is a chance the experiment lacks the power to detect what could be a rather small magnitudinal change in expression between conditions.

[bionewbie]

My experiment design consists of a control and 2 mutation types. I tried to create DESeq2 conditions MA vs WT and MD vs WT. I am getting upregulated and downregulated genes but it does not consist of the genes that I am interested in. The genes that I am interested in had a log fold change value between 0 and -1 and an adjusted p-value of 1.

[jared.andrews07]

Yes, that happens sometimes. The results are the results - massaging them to just show what you want are not going to help you determine what's going on in the actual biology. You may want to determine what is interesting about the genes that are differentially expressed or discuss alternative results with your PI.

Answer 2

[German.M.Demidov]

I can't say that it is frequent for me to see a p-value of 1 (as well as p-value of 0). It can be that the read counts are too small or p-value is 1 after multiple test correction. Are you sure that you refer to p-value and not to q-value? Are you sure that you used a double-sided test (so you have not mis-specified in which group the expression expected to be elevated)? Could it be that your cases/controls come from different batches and the batch correction was overly aggressive? Can it be that your case/control status somewhat leaked into your design matrix of predictors?

I agree with Jared Andrews - every failed experiment requires a careful autopsy, and without knowing the details it is difficult to perform it.

[bionewbie]

My experiment design consists of a control and 2 mutation types. I tried to create DESeq2 conditions MA vs WT and MD vs WT. I am getting upregulated and downregulated genes but it does not consist of the genes that I am interested in. The genes that I am interested in had a log fold change value between 0 and -1 and an adjusted p-value of 1

[German.M.Demidov]

How many biological replicates are in MA/MD/WT groups? How many technical? I will think about your design matrix, if you specified your outcomes there you may get a non-significant p-value in the end...

[bionewbie]

res <- results(dds, alpha = 0.05, contrast = c("cell_line","MA","WT"), lfcThreshold=1) summary(res)

out of 21991 with nonzero total read count

adjusted p-value < 0.05

LFC > 1.00 (up) : 274, 1.2%

LFC < -1.00 (down) : 606, 2.8%

Similarly for the other condition,

res <- results(dds, alpha = 0.05, contrast = c("cell_line","MD","WT"), lfcThreshold=1) summary(res)

I had 3 biological replicates for each sample condition.