estimateDispersions warning message when using data imported from Seurat v3

[benjytan88]

Hi. I am using Seurat v3 for my analysis and I imported the data manually into Monocle as per the instructions here: https://github.com/cole-trapnell-lab/monocle-release/issues/262#issuecomment-465749521

However, when I run the estimateDispersions function, I received a warning message as shown below:

NaNs produced step size truncated due to divergence glm.fit: algorithm did not converge Dispersion fit did not converge.Removing 10 outliers

The data looks OK as I was able to run the next steps but I am not sure what this warning message is trying to inform me. Is there anything faulty with my data? Anyone face this problem before and can you explain it to me?

[mragsac]

Hello, I ran into a similar error after importing my Seurat v3 object using the instructions at #262 .

Warning message in log(ifelse(y == 0, 1, y/mu)): “NaNs produced”Warning message: “step size truncated due to divergence”Removing 110 outliers

I was wondering if you ran into any issues with the differentialGeneTest() method.

[roosheelpatel]

I am also experiencing this issue. Any insights would be greatly appreciated!

[Sandradehaan]

Hi! did someone solve this issue already by any chance?? Running into the same problem..

[mb84]

Me too, i've the same issue when loading "raw data" (counts) slot. By using the "data" slot i don't have any problem.

[Sandradehaan]

I am also having this issue, anyone knows how to solve it?

[Rvirgenslane]

I had the same problem and got a fix.

You need to add a model to fit for dispersion estimation:

HSMM<-estimateDispersions(HSMM, modelFormulaStr=“~Condition”)

Default model is “~1”

[Bernadetadad]

Getting same error:

“glm.fit: algorithm did not converge”Warning message in parametricDispersionFit(disp_table, verbose):
“Dispersion fit did not converge.”Removing 177 outliers
Warning message in log(ifelse(y == 0, 1, y/mu)):
“NaNs produced”Warning message:
“step size truncated due to divergence”Warning message in log(ifelse(y == 0, 1, y/mu))

Not sure Rvirgenslane answer solves the problem if you want to calculate Dispersions not by specified variable.

[penglbio]

I had the same problem and got a fix. When you create the CDS_object, You can't use sparse matrix. Use as.matrix()like the following:

my_cds <- newCellDataSet(as.matrix(gbm), phenoData = new("AnnotatedDataFrame", data = pd), featureData = new("AnnotatedDataFrame", data =fd), lowerDetectionLimit = 0.1 ) my_cds <- estimateSizeFactors(my_cds) my_cds <- estimateDispersions(my_cds,cores=32)

[lisahnglai]

hello，I had a problem like this： cds <- estimateDispersions(cds) Removing 17 outliers There were 50 or more warnings (use warnings() to see the first 50) anyone knows how to solve it?