DESeq2 LRT not returning genes
1
0
Entering edit mode
meeta.mistry ▴ 30
@meetamistry-7355
Last seen 4 weeks ago
United States

I have encountered something really confusing when I did my LRT analyses. I'm interested in gene P's effect on cellular immune reactions. In experiments, we treated the WT and P KO cells with either normal media (control), or drug A, or drug B to induce immune responses. For each treatment, we treat the cells for 6 hours and 24 hours.

In these experiments, we find a change in metabolites in WT cells treated with drug A for 6 hours, and also in WT cells treated with drug B for 24 hours. We didn't see this metabolic change in P KO cells. To understand what is going on, we did RNAseq experiments on these WT and KO cells.

Biological question: for a given treatment, which genes are differentially expressed in WT, that do not change expression in P KO (or change differently). Basically what impact does P KO have the transcriptional pattern across treatment (untreated/media, 6h, 24h). We expect some genes given the differences in metabolites observed (described above).

If we divide the data by treatment, my metadata looks like this:

For this purpose, I used the LRT with the interaction term to understand gene P’s effect on the treatment. But when I do this, all the genes show padj =1, which doesn’t make sense. So I wonder if there is anything wrong in my code? Or is there other better way to analyse this data?


dds_geno <- DESeqDataSetFromMatrix(data, colData = meta_A, design = ~ genotype + treatment + genotype:treatment)
dds_geno_lrt <- DESeq(dds_geno, test="LRT", reduced = ~ genotype + treatment)


Meeta

LRT linearmodels DESeq2 • 108 views
0
Entering edit mode
@mikelove
Last seen 2 hours ago
United States

This just means that the differences in treatment effect across genotype were too small to detect with your sample size.

If you have genes you believe should have had differential treatment effect, try plotting them with plotCounts.