Error in matrix(0, nrow = length(eids), ncol = ncol(expTPM)): "non-numeric matrix extent"

[Darlingfuer]

Hi,

I have been trying to run CellNet locally on my Mac. But there's error "non-numeric matrix extent" at the step "Estimation of expression levels".

Here's my code:
expList<-cn_salmon(stQuery, refDir="ref/", salmonIndex=iFileMouse,salmonPath=pathToSalmon)

When I trace back, it's the line "ansTPM<-matrix(0, nrow=length(eids), ncol=ncol(expTPM))" at function "gene_expr_sum" at "cn_salmon".

I'm not familiar with R, so would you please help with this?

[Darlingfuer]

I've checked the class of expTPM. After the line "expTPM<-expDatList[['TPM']]", class(expTPM) is matrix. But after the line "expTPM<-expTPM[sameProbes,]", class(expTPM) is numeric. Then when it gets to "ncol=ncol(expTPM)", the "ncol" is NULL.

[pcahan1]

How many fastqs are you trying to process?

[Darlingfuer]

@pcahan1 Sorry, I realized I was trying to test with only one fastq file.
But when I tried to test with three fastq files, I got another error at cn_salmon:

./salmonRes_FKO_mESC/quant.sf
./salmonRes_WT/quant.sf
./salmonRes_FKO_diff/quant.sf
Error in geneIndexList[[i]] : subscript out of bounds

Actually I didn't got any error when I test on my Mac. But I couldn't figure out what's going on when I tried to test it on server. I used the latest version of salmon and build the index myself.

[Darlingfuer]

I wonder where I can download the fasta file you used to build index: Mus_musculus.GRCm38.cdna.ncrna.exo.fa

[pcahan1]

Hi, We are working on this and hope to have a fix very soon. Will post here as soon as we can. Thanks.

[pcahan1]

The latter issue

./salmonRes_FKO_mESC/quant.sf
./salmonRes_WT/quant.sf
./salmonRes_FKO_diff/quant.sf
Error in geneIndexList[[i]] : subscript out of bounds

is likely due to using a different index that what we have provided.

We have fixed the earlier error in gene_expr_sum that pops up when the pipeline is applied to one query sample.

Please let us know if you have any more issues. Thanks!

[Darlingfuer]

@pcahan1 Thanks!
I then used cdna and ncrna files from Ensembl to rebuild the index with the latest salmon software and it works throughout the cn_salmon function.
I noticed that the index file that you provided (that is "salmon.index.mouse.122116") doesn't work well with the latest version of salmon(0.8.2), which would raise a error so that we need to rebuild the index with latest version of salmon.
However, I got another problem at the "Reconstruction of GRNs" step when using the "cn_make_grn" function:
esc : 5
liver : 3
neuron : 5
VE : 1
Number of samples per CT: 1
Error in expDat[, rownames(stGRN)] : incorrect number of dimensions
Calls: cn_make_grn
In addition: Warning message:
In if (is.na(tfs)) { :
the condition has length > 1 and only the first element will be used

I was trying to use my own data to train a new CellNet, btw. The training metadata table is in ".rda" format in your paper. I just read my csv file and save as ".rda" in R and then use it for cn_make_grn function together with the output of cn_salmon function. Is there anything wrong with this conversion ? I hope that I describe my situation clearly.

[Darlingfuer]

The output file of cn_salmon is a list, while the parameter for cn_make_grn need to be a matrix ? Step 13-16 in your Nature Protocol paper doesn't describe how to build a new CellNet in details. So far, I have just finished collecting fast files, creating a training metadata table in "csv" format and run through the cn_salmon function...

[pcahan1]

Hi, I am going to move this last error

Error in expDat[, rownames(stGRN)] : incorrect number of dimensions

to another issue and close this one as it is not related to the original bug. I cannot recommend using so few samples per cell type or tissue. In the original CellNet paper, the minimum was around 60 samples per cell type, and in the rna-seq we have gone as low as 30-40.