Doubt about model in paired samples (intercept ?)
1
0
Entering edit mode
picasa1983 • 0
@picasa1983-14806
Last seen 19 hours ago
USA

I have 3 donors, and each donor has 2 samples (KO and WT). The matrix is:

Donor_ID    Condition
1   WT
1   KO
2   WT
2   KO
3   WT
3   KO


I am interested in looking at differentially expressed genes (DEG) between KO and WT while correcting for donor variability.

I have doubts about my model, especially regarding the intercept. Should it be:

dds <- DESeqDataSetFromMatrix(data, colData = meta, design = ~ Donor_ID + Condition)
res <- results(dds, name = "Condition_KO_vs_WT")


or

dds <- DESeqDataSetFromMatrix(data, colData = meta, design = ~ 0 + Donor_ID + Condition)
res <- results(dds, name = "ConditionKO")


because I get different results after using lfcShrink.

DESeq2 • 413 views
1
Entering edit mode
@mikelove
Last seen 11 hours ago
United States

Take a look at ExploreModelMatrix to help understand what coefficients mean in a linear model.

The top one is the KO vs WT comparison.

0
Entering edit mode

1) I looked at ExploreModelMatrix, but it didn't help me much. For instance, in the first model with an intercept, I don't see any mention of the coefficient Condition_KO_vs_WT (which I can also get from resultsNames(dds)). Instead, I see the coefficient ConditionKO, which is a bit confusing. Why is that?

2) If the first model represents the KO vs WT comparison, how can we interpret the second model without an intercept?

0
Entering edit mode

I have to restrict my time on the support site for software related questions.

For questions about statistical designs and analysis choices, I'd recommend consulting with a local statistician, or anyone familiar with linear models in R. DESeq2 uses the same linear modeling framework as basic linear models implemented in R, e.g. lm and model.matrix.

Traffic: 693 users visited in the last hour
FAQ
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.