Assessing the influence of covariates on RNA-seq in DESeq2
1
0
Entering edit mode
rmash • 0
@rmash-21927
Last seen 14 months ago

I have an experimental design that I did in 2 blocks (1 week apart). I'm struggling to understand how DESeq2 deals with covariates.

I checked a PCA and there is no clustering of blocks.

I've run a model

dds <- DESeqDataSetFromMatrix(countData = genes.matrix,
colData = sample_info,
design = ~ Group + block)


where block is coded as -0.5 and +0.5, with the idea that this should mean center the effects? But is this true? I've also coded as "B1" and "B2" but this has no impact on what DEGs are expressed.

How do I tell if block is having a significant influence on the overall results?

Would one interpret a significant result in group as occurring over and above variation in block? as you would in a regular linear model?

Any help appreciated!

deseq2 rna-sequencing covariates glm • 152 views
0
Entering edit mode
@mikelove
Last seen 2 hours ago
United States

If you add a nuisance variable, it will account for variation if it exists per gene, so usually the approach would be ~ batch + condition (this is in the vignette for example).

You should just keep batch as a factor, it will "average" over the levels of batch when estimating the condition effect. You don't need to code with the -.5 and +.5 to accomplish this.