Question

weights must be finite positive value

0

Entering edit mode

nonCodingGene ▴ 10

@noncodinggene-7018

Last seen 6.4 years ago

I'm using edgeR and an error for which I have not found a solution just appeared.

design <- model.matrix(~Cell.type, data = colData(summ.exp_norm))
dge$weights <- zinb.weights
dge <- estimateDisp(dge, design)

Error in .compressWeights(y, weights) :  weights must be finite positive values

As you can se no weight is out the rang 0-1 or infinite.

> sum(is.infinite(dge$weights))
[1] 0

Where can the error be?

Thanks

edger estimatedispersions • 1.7k views

ADD COMMENT • link updated 6.4 years ago by Aaron Lun ★ 28k • written 6.4 years ago by nonCodingGene ▴ 10

0

Entering edit mode

Why do you think is.infinite will tell you if any values are less than zero? Or for that matter, within the range 0-1?

ADD REPLY • link 6.4 years ago James W. MacDonald 67k

0

Entering edit mode

Sorry, I forgot to show that also tested for max(dge$weights) and min(dge$weights), and values where 1 and 0.

ADD REPLY • link 6.4 years ago nonCodingGene ▴ 10

0

Entering edit mode

Well, the test is

check.range <- suppressWarnings(range(weights))
    if (any(is.na(check.range)) || check.range[1] <= 0) {
        stop("weights must be finite positive values")
    }

So really it's saying you have either negative or NA values. This comes after running makeCompressedMatrix, so you might see if there are any NA values after that step with your weights matrix.

ADD REPLY • link 6.4 years ago James W. MacDonald 67k

0

Entering edit mode

sumis.na(dge$weights))
outputs 0

ADD REPLY • link 6.4 years ago nonCodingGene ▴ 10

score 0 · Answer 1 · 2018-07-18

0

Entering edit mode

Aaron Lun ★ 28k

@alun

Last seen 2 hours ago

The city by the bay

Pretty simple, really. If min(dge$weights) is zero, that's not positive.

ADD COMMENT • link 6.4 years ago Aaron Lun ★ 28k

0

Entering edit mode

Oh yeah, <= 0 ...

ADD REPLY • link 6.4 years ago James W. MacDonald 67k

0

Entering edit mode

What can be the source of this and how can I solve it?

I'm using edgeR for scRNA-seq, in order to manage droputs I use zinbwave

Just in case I've checked whether one condition has a gene for which its counts are all 0, and this does not happens.

This is how I calculate weights:

 zinb <- zinbFit(summ.exp_norm, K=2, epsilon=1000)
 zinb.weights <- computeObservationalWeights(zinb, assay(summ.exp_norm))