extract the samples with CNV

[QSong-github]

Thanks for the fantastic tool! I am currently using it on my single cell data.

However, I found some inconsistencies when I try to extract the cells with CNV.

First, I used this output "bin_mat=binarizeMatrix(redu,normal,tumor,0.8); cells1<-apply(bin_mat,1, sum)" . I suppose the cells1 with higher value are cells with CNV.

On the other hand, I also tried to use the heatmap generated by "plot1<plotBinaryMat(bin_mat,patients,normal,tumor,patient=patients); CNVs= cutree(plot1$tree_col,k=3)". Then I took the cluster that showed with CNVs in heatmap. Surprisingly, the cells differed greatly with the cells1 that I got above.

I feel so confused about it. Could you give some information? Which one is right or how to get the cells with CNV?

Thanks a lot!

[soerenmueller]

Hey, thanks for using CONICSmat!
First idea I have that could be the issue is that binarizeMatrix will produce NAs for cells that have a posterior between 0.2 and 0.8 based on the threshold parameter (0.8). Therefore, most likely cells1<-apply(bin_mat,1, sum) will have entries which are NA. You could either use cell1->apply(bin_mat,1, sum,na.rm=T). If you do that, could you run
intersect (cells1,cells2), where cells2 are the cells you selected by cutting the tree?

A second way to avoid this would be:
bin_mat=binarizeMatrix(redu,normal,tumor,0.5). This will threshold the posterior probability on 0.5 and therefore no NAs will be generated.

Let me know if this helped.
Thanks,
Soren

[soerenmueller]

If you update your CONICSmat version (you can just re-install it from git), I changed the code for plotting the binary matrix. On the bottom right it now indicates which cluster has which ID and returns a cluster assignment for each cell. The order of the clusters in the plot might have been an issue as well. Pheatmap does not order them sequentially.

plot1=plotBinaryMat(bin_mat,patients,normal,tumor,patient="MGH97")

which(plot1==3)
    97_P3_A07     97_P3_C10     97_P3_B02     97_P3_B01     97_P3_H10     97_P5_C04     97_P6_H04     97_P5_C01     97_P6_A11     97_P6_F10 
           24            44            58            65           105           121           125           147           162           223 
 MGH97_P7_G11  MGH97_P7_E01 MGH97_P10_H07 MGH97_P10_B10  MGH97_P8_B05 
          276           345           438           452           535

[QSong-github]

@soerenmueller
Thanks for your prompt reply. Actually my code removed the NAs, i.e.
"extract.cells <- apply(bin_mat,1,function(i){sum(as.numeric(na.omit(i)))})" ; so I don't think it's the NAs issue.
I will re-install the package to update the function and check if it works now. Thanks a lot and will let you know then.

[soerenmueller]

@wasqqdyx
I checked again today with fresh eyes and actually there was a bug in the plotBinaryMat function. Thanks for pointing me to this! The names of cells were incorrectly assigned. I fixed the bug in the current version.

which(cells1==0)
    97_P3_H11     97_P3_H03     97_P3_A11     97_P3_F01     97_P5_A04     97_P5_B06     97_P5_E07     97_P5_D04     97_P6_E04     97_P5_G10 
           24            44            58            65           105           121           125           147           162           223 
 MGH97_P7_C11  MGH97_P7_D08 MGH97_P10_G01  MGH97_P8_D10  MGH97_P9_C08 
          276           345           438           452           535 
> which(plot1==3)
    97_P3_H11     97_P3_H03     97_P3_A11     97_P3_F01     97_P5_A04     97_P5_B06     97_P5_E07     97_P5_D04     97_P6_E04     97_P5_G10 
           24            44            58            65           105           121           125           147           162           223 
 MGH97_P7_C11  MGH97_P7_D08 MGH97_P10_G01  MGH97_P8_D10  MGH97_P9_C08 
          276           345           438           452           535 

[soerenmueller]

I'm closing this thread, thanks for pointing me to the issue.