Dear all,

I'm involved in several projects where RNAseq data are collected in time at multiple time points on the same mouse or patient and the goal is to detect DE genes over time or at specific time points. I have checked the DESeq2 vignette (section: Time-series experiments) but it is not clear to me how exactly DESeq2 models the correlation between the repeated measurements collected on the same unit. Could you please refer me to the right documentation to read about it?

Thank you in advance for your help.

Kind regards,

Roula

Dear Wolfgang,

Thank you very much for your reply. I’m familiar with the GLM theory and testing in such a case. Typically when data are collected repeatedly from the same subject i.e. mouse/human the correlation between the repeated measurements should not be ignored. One approach could be to use generalized mixed effects models or the generalized estimating equation approach. And this is what I wanted to understand about DESeq2.

If I understand it correctly, for time course experiments DESeq2 treats all samples as independent. This is fine if at each time point a different group of mice/humans is sequenced. However, if blood/tissue is taken from the same mouse/human at several points in time then the potential correlation between the time-points should not be ignored. In my opinion this is not clearly stated in the vignette and this could be very useful for the users to know.

Kind regards,

Roula

If you want to make within-individual comparisons across groups of individuals, you can use this approach to control for the correlation of samples from the same individual:

https://bioconductor.org/packages/release/bioc/vignettes/DESeq2/inst/doc/DESeq2.html#group-specific-condition-effects-individuals-nested-within-groups

However, a direct comparison across groups of individuals (not within-individual), while controlling for repeated measures from individuals is not possible with fixed effects modeling available with DESeq2. Here the duplicateCorrelation() function and the limma-voom methods for RNA-seq can be used instead.