Interpreting likelihood ratio tests using DESeq2

Question

[jdhindsa1999]

Hello,

I am hoping to receive some guidance on a longitudinal RNA-sequencing project I've conducted. I have sequences at multiple ages across 3 different disease models in addition to control. I am running a likelihood ratio test in DESeq2 with the full and reduced models as specified:

full: ~ genotype + age
reduced: ~ age

I realize that I am testing for the effect of genotype, controlling for age, one gene expression with this test. However, I'm unsure how to interpret the DESeq results and particularly the log fold change result. The padj values are simply telling me whether my full model is a better model to capture deviance of gene expression, without taking into account log2fold change differences. Does this mean I could have significant DEGs for which the log fold change in expression is actually 0?

Answer 1

[rpolicastro]

The Log2 FCs are not actually directly related to the LRT test p-values, but rather the result of a comparison between two groups (controllable by the names argument). For your purposes just ignore them for now.

The p-values in your case will show genes where the additional coefficient (genotype) produces an increase in log-likelihood that is larger than would be expected if the coefficient was 0 (functionally having no effect).

[1769mkc]

how the Log2 FCs is calculated in case of single factor or condition lets say three groups, in LRT it would run as reduced model and the it compares gene across all the 3 condition it is explained that the compares the full vs reduced model to report the significant genes. So in case of LRT when i run the reduced model which sample or group becomes my reference?