Hi, I would like to perform a differential variance analysis where, instead of modeling expression~(age), I model variance ~ age.
I have pseudo bulk raw counts per sample (n = 300). For differential expression analysis, I used DESeq2, as the data is best modeled with a negative binomial (NB) distribution. However, I am unsure how to approach differential variance analysis.
If the data had a linear mean-variance relationship, I would extract residuals from lm (expression ~ (age), compute the squared residuals, and then model squared residuals ~ age using linear models. However, the variance in my data is not constant, even after attempting inverse rank transformation (heteroskedasticity remains).
i assume i have to model the data using NB, but I am unsure how to do this, and if i can implement DESeq2. Would a variance-stabilizing transformation (VST) followed by linear modeling be an appropriate approach?
I would appreciate any suggestions on best way to do so
Thanks Michael for your answer,
i didnt mean to use Deseq2 directly for the variance testing, but more like asking if using the (VST) data is valid for this tests , if you have in mind any reference of someone doing so, would love to take a look .
also if you have a package in mind that you think is the standard for DV testing. i would appreciate your guidance cause i have looked at some tools and there benchmarks but most of them don't seem to be user friendly or heavily utilized
https://pmc.ncbi.nlm.nih.gov/articles/PMC10544356/
thanks agaain