Dispersion shrinkage in DESeq2: LOESS instead of negative binomial based GLM?

Question

[CY]

DESeq2 uses GLM with negative binomial distibution to model the regressor for variance shrinkage. It is obvious that the expression variance is different for gene with different expression level and a ordinary linear model is not suited. I am wonder if it is possible to model this regressor using LOESS instead of GLM?

[ATpoint]

You can ask the maintainer for an experts opinion at BioC (https://support.bioconductor.org/) but I would avoid the term obvious unless you are 100% sure what you are talking about and have experts knowledge on why things are modeled/used as they are.