@koenvandenberge @drisso Sorry to quote a closed issue #33 but I'm trying to understand what would be the most appropriate choice when computing DE genes.
I have a 10x dataset where I used the zinbwave model to compute the W matrix. I have 7 patients, 2 biological conditions(treated vs controls), and within the treated condition 2 different drugs. (i know, it's a hell of a mess). the aim was to correct for the known variability in the experimental design, and to account as well for any other source of variation, to cluster the cells based on their top genes ignoring the fact that the patients are heterogeneous and belong to 2 different groups. i tested several zinbwave params and ended up with the following design: X=~patient.id+drug.treatment, k=10.
then used the W matrix to do clustering. (as suggested in the vignette)

now I need to perform differential expression WITHIN cluster (so it's n differential expression testings, n= number of clusters), contrasting on the "condition" level (treat vs untreat). this is a slightly different problem from the first task and I am wondering what is the appropriate choice for the weights estimation using zinbwave. given that for each cluster I have cells from 2 conditions and all the patients are represented (not homogeneously but that's the inherent variability of the experiment I guess) should I estimate the weights with zinbwave(zinb, K=0, X=~patient.id+condition, epsilon=1e12) and use the same design with Deseq for the diffexp test? And should I really use K=0?

thank you!