Hi,
I have time-series data (in order to measure the half-life of RNA degradation) under different conditions. To test any genes react in a condition-specific manner over time, I could test the interaction term with a LRT test, e.g. ~ condition + time + condition:time as the full model and ~ condition + time as the reduced model), similar to the example stated here:
http://master.bioconductor.org/packages/release/workflows/vignettes/rnaseqGene/inst/doc/rnaseqGene.html#time-course-experiments
Alternatively, I could also calculate the half-life of each gene based on their time-series normalized expression values (like the HalfLivesBlock() in DRUID code) and then model the half-life (instead of gene expression) simply as half_life ~ condition.
My question is, if I would take the second method, I would need to use the normalized expression values (e.g. TPM), not raw reads, to calculate half-life. After that, can I still be able to use DEseq2 to model the differential half-life? I doubt it, since half-life may not fit the negative binomial distribution like raw reads count. DEseq2 takes raw count because it will internally model its dispersion (mean ~ deviation relationship) and model it in NB distribution. If I convert the time-series expression to a new variable (e.g. half-life, progression rate etc.), is my new data still eligible for DEseq analysis?
Please advise. Thanks
