Hello, I'm new to this forum and to RNA-seq, and I've been having problems with my data analysis (I use Partek Flow). I'm trying to compared data from animals that are WT and Het for gene X and that were treated with different concentrations of a hormone, so I know for a fact that at least gene X should be differentially expressed (the treatment would not change this at all). But when I look at my data, I don't see any genes that have significant FDR. There are a bunch of genes that have p-value < 0.01 and fold change greater than 1.5 (neg. or pos.), but the FDR is much higher than 0.1. In one condition, I can get a list of genes if I filter it with FDR < 0.25, in the other, the FDR is 0.686534 for all genes, regardless if the p-value is 0.01 or 0.001, or something in between. Even gene X has a high FDR, and I know for a fact it is differentially expressed between the WT and Hets. I've actually used the sample samples in qPCR and gene X was significantly lower in the Hets, at it should be. So why do I get such high FDR? Is there something wrong with the data? Everything seemed fine (good RNA quality, good QC for the run, good pre and post-alignment QCs, etc.) until now. Am I doing something wrong in the analysis? There's not much I can do wrong though, since I don't have to code anything, and Partek has a pretty standard pipeline, but perhaps I can change something with the settings or if there's a specific test I should be watching out for? This is the last experiment standing between me and my Ph.D. defense, so I would like to get it done quickly. I have more of the samples I used in the library prep, and other animals I could use, so I could start from scratch, but I just wanted to know if maybe I can adjust something in the analysis instead of starting it all again. Thank you!
Hello, how many replicates do you have? Please add code (if possible, not sure with Partek). Did you do QC checks e.g. with PCA to see how data cluster? Not much to say without more details I fear.
I just want to add that a histogram of p-values would be helpful in addition to the above too, since FDR correction works by building a cutoff based on the distribution of p-values.
Please see How to add images to a Biostars post to add your images properly. You need the direct link to the image or the HTML embed code, not the link to the webpage that has the image embedded (which is what you have used here)
From looking at the histogram I wouldn't expect a large number of significant values, but it's likely there should be some. Can you put your code in the post that generated this data?
Hello, I have 3 biological replicates for each condition for a total of 18 samples: 3 wt 3 het concentration X 3 wt 3 het concentration Y 3 wt 3 het concentration Z
I'm filtering the data out before doing the GSA, so for example, I'll run the GSA on just concentration X (WT and Hets), then do the GSA on WT X and Hets X, Y and Z, and so forth.
When I do the PCA, the groups usually cluster together, except for condition Z that might have one sample that doesn't cluster all that well with the others in the group. But concentration Z actually has the lowest FDR.
I don't have the code, but here are the details for the GSA, if it helps:
Factors: Condition Comparisons: WT X vs. Het X (Condition) Low-value filter: Lowest average coverage = 1.0 FDR step-up: true Storey q-value: false Bonferroni: false Model selection criterion: AICc Enable multimodel approach: Yes Use only reliable estimation results: Yes Model types configuration: Default (Min error degrees of freedom: 6, Lognormal with shrinkage, p-value type: F)
Thank you!