DESeq2- preliminary study no replicates
1
0
Entering edit mode
@yonatan-amzaleg-19022
Last seen 17 months ago
Los Angeles, CA, University of Southern…

Hello,

First off, I appreciate everyone reading this and providing insights. I'm aware what I'm asking is not statistically sound by any stretch of the imagination, however the current data I have is three samples (one replicate each). The conditions are such that two of the samples share a condition (8 weeks vs 3 days) and two of them share another condition (hypoxia vs normoxia).

I was able to make a PCA plot and heat map of these samples by doing a DESeqDataSetFromMatrix with the design being the thing that at least two of the samples share in common (I did it with both and produced similar PCA plots).

Before I can proceed with running the replicates, I was also charged with doing a preliminary DEG with the samples I have now. However, the comparison I need is not Hypoxia vs Normoxia or 8 weeks vs 3 days , but rather one specific sample against the other.

so I repeated the DESeq2 analysis this time changing the design to Samples and when running the DESeq command I got the error

dds <- DESeq(dds)

estimating size factors
estimating dispersions
Error in checkForExperimentalReplicates(object, modelMatrix) :

The design matrix has the same number of samples and coefficients to
fit,   so estimation of dispersion is not possible. Treating samples
as replicates was deprecated in v1.20 and no longer supported since
v1.22.


Is there anyway to avoid this error? For instance, do I have to reinstall the older version of DESeq2 v1.20? Does that mean that I would have to re run some of my older stuff to fit to this new version? I get the importance of not having it available any more, but this is more about a preliminary experiment to see if some genes where I anticipate Differential expression actually have differences.

Yonatan

deseq2 replicates error DESeqDataSetFromMatrix • 2.5k views
1
Entering edit mode
@mikelove
Last seen 40 minutes ago
United States

I’d recommend that you compute the vst() of the dataset and look at the large LFCs (difference between the two VST samples). I don’t think the pvalues were useful at all, hence the deprecation. I think the best we can do is output shrunken LFC (which here you can get by looking at difference between VST samples).

0
Entering edit mode

Thank you so much for your response.

I'm a little confused as to how to perform the vst function without having run the DESeq function first? Should I run it with a design that works and then transform the data?

Yonatan

1
Entering edit mode

You can use ~1 to build the dataset.