Hi,

I am using DESeq2 to analyze a (short) time series gene expression dataset. Does anybody know what is the advantage, if any, of using an LRT (likelihood ratio test) for differential expression analysis across all time points versus a Wald test separately for each time point?

I have tried both but I do not see much qualitative differences.

Thank you in advance.

Denise

The advantage to using the LRT is that it's an omnibus test that tells you if there are changes at any time point. If you have lots of time points you could have any number of Wald tests you would have to do. The alternative is to do the LRT, get the significant genes, and then use e.g., k-means clustering to identify groups of genes with different patterns of expression. You can then make high level statements about those groups, and maybe do gene set testing of some sort to identify likely pathways that are being affected in a time-dependent manner.

Think of it as inspecting the forest versus looking at individual trees. In general the higher level view provides (IMO) a more identifiable signal.