Strange BCV plot for 10x data #41
Description
Hi
I have been trying ZINB-Wave for generating weights for edgeR for a 10x dataset and I have got some results that look a bit strange. Here is what I have been doing so far:
zinb_model <- zinbFit(counts(sce),
X = model.matrix(~ Cluster, data = colData(sce)),
epsilon = 1e12,
verbose = TRUE)
zinb_weights <- computeObservationalWeights(zinb_model,
as.matrix(counts(sce)))
n_clusts <- length(unique(colData(sce)$Cluster))
design <- model.matrix(~ 0 + colData(sce)$Cluster)
colnames(design) <- paste0("C", seq_len(n_clusts) - 1)
dge <- DGEList(counts(sce))
dge <- calcNormFactors(dge)
dge$weights <- zinb_weights
dge <- estimateDisp(dge, design)
de_fit <- glmFit(dge, design)
plotBCV(dge)Which gives this (strange looking) BCV plot:
The regular BCV plot (without ZINB-WaVE) weights looks ok:
Just wondering if this is normal, I've done something wrong or perhaps the dataset isn't a good fit. Also what effect this is likely to have on downstream DE results?
Thanks