Error in preprocessing

[smanne07]

Hi Hannah,
I have succesfully the cicero pipeline on 10x generated data but for only one of the samples i have the following error:

Error in if (any(i < 0L)) { : missing value where TRUE/FALSE needed Calls: [<- -> [<- -> [<- -> [<- -> int2i In addition: Warning message: In int2i(as.integer(i), n) : NAs introduced by coercion to integer range

Does this error point to anything with how the data is for this sample?I tried to check from where this error is raised and i could see that this part of the preprocess/reduceDimension functions,but i cannot figure out why

Can you help in troubleshooting this?

Thank you
Sasi

[hpliner]

There've been a couple of reports of this problem but I haven't been able to reproduce it with the publicly available 10x datasets. A couple of questions: is this human data? Is there any possibility of NAs, Infs or NaNs in your expression matrix?

[smanne07]

This is mouse data,i will check if and how the NAs were introduced in this dataset

[hpliner]

Ok, let me know if possible. Also happy to debug myself if you can share the data

[smanne07]

Hi Hannah,
Looking in to the data it looks it is an issue similar to sdparekh/zUMIs#105

This is happenning when we have data from larger number of cells(In my case 14000 failed against 10000 which was succesfull)

https://bitbucket.org/hrue/r-inla/issues/1/logical-indexing-for-large-matrices-fails

Thank you
Sasi

[smanne07]

traceback()
7: int2i(as.integer(i), n)
6: [<-(x, i, value = value)
5: [<-(x, i, value = value)
4: [<-(*tmp*, ncounts != 0, value = 1)
3: [<-(*tmp*, ncounts != 0, value = 1)
2: normalize_expr_data(cds, norm_method, pseudo_expr)
1: reduceDimension(input_cds, max_components = 2, num_dim = 6, reduction_method = "tSNE",
norm_method = "none")

normalize_expr_data for binomialff family

FM <- exprs(input_cds)
ncounts <- FM > 0
ncounts[ncounts != 0] <- 1

Is causing issues with larger matrices.

[hpliner]

Thanks for the traceback and info. I just pushed a fix to monocle3, so if you rerun the installation described here: http://cole-trapnell-lab.github.io/monocle-release/monocle3/ It should work now!

[smanne07]

hi @hpliner ,
I was trying Cicero with the latest monocle3 installed and i am still facing the same issue with the error

Error in if (any(i < 0L)) { : missing value where TRUE/FALSE needed Calls: [<- -> [<- -> [<- -> [<- -> int2i In addition: Warning message: In int2i(as.integer(i), n) : NAs introduced by coercion to integer range

What is the current recommendation for running Cicero on large datasets?Which monocle version?

Thank you
Sasi

[huidongchen]

Hi

Thanks for creating this amazing tool. Unfortunately i'm also having the same issue.

Warning message in int2i(as.integer(i), n):
"NAs introduced by coercion to integer range"
Error in if (any(i < 0L)) {: missing value where TRUE/FALSE needed
Traceback:

1. reduceDimension(input_cds, max_components = 2, num_dim = 15, 
 .     reduction_method = "tSNE", norm_method = "none")
2. normalize_expr_data(cds, norm_method, pseudo_expr)
3. `[<-`(`*tmp*`, ncounts != 0, value = 1)
4. `[<-`(`*tmp*`, ncounts != 0, value = 1)
5. as(.TM.repl.i.mat(as(x, "TsparseMatrix"), i = i, value = value), 
 .     "CsparseMatrix")
6. .class1(object)
7. .TM.repl.i.mat(as(x, "TsparseMatrix"), i = i, value = value)
8. `[<-`(x, i, value = value)
9. `[<-`(x, i, value = value)
10. int2i(as.integer(i), n)

In my case I have ~12,000 cells

[hpliner]

Cicero should now be up to date with the new Monocle3 beta (https://cole-trapnell-lab.github.io/monocle3/) on it's monocle3 branch. Can you check out the latest instructions here https://cole-trapnell-lab.github.io/cicero-release/docs_m3/ and let me know if that solves your issue?