error in classify tumor cell and nontumor cell?

[zhumengyan]

Hi,
Thanks for your software. I want to use CONICSmat to classify tumor and nontumor cells in a 10X dataset. AND I come across two probelms.
1. Compared to tumor cell markers, I'm surprised that CONICSmat result may be reversed.
2. The color of Chromosome Heatmap is very light.
I really appreciate your positive reponse.Thank you very much!

Best regards
Mengyan Zhu

[biobug16]

Hi Mengyan,
Even I am facing the same problem with my data. Did you figure-out this issue?
In case you have, will you mind to share your experience about how did you manage to get it solved?

Thanks

[dstueckm]

I'm getting the same issue. I'm looking for specific chromosomal deletions as tumour markers, but the histogram used to detect tumour/normal does not keep the order of clusters consistent (in terms of high expression and low expression clusters) between samples. Therefore, using normal = which(hi==1) as shown in the tutorial does not always work. Is there any way to specify a deletion event rather than a generic CNV event?

[biobug16]

Hi dsteuckm,
Its not always "normal = which(hi==1)", you need to check the figure you generate after the step
candRegions=rownames(lrbic)[which(lrbic[,"BIC difference"]>200 & lrbic[,"LRT adj. p-val"]<0.01)]
That figure will show you which cluster does not harbor any of the copy number in your dataset and pick that one as normal.

I hope this is the answer you are looking for.

[dstueckm]

Hi @biobug16,

Thanks for the reply. I should have clarified; I do understand that you can determine which clusters are normal, my point was more from an automation perspective. If I am running a large number of samples, and the histogram produced does not necessarily have any "order" in terms of which CNVs occur from left to right, it will require visual inspection each time. It would be helpful to have a function with a readout of which clusters contain which CNVs, so we could then subset the histogram in that way.

[diazlab]

Hi dstueckm, It is a good question and unfortunately the current code does not do that. It is an issue, as you say, when running a large number of samples since we then have to manually go back and look at the plots. The issue could be addressed with some additional code. We are somewhat swamped at the moment. If you were to write some code for that it would be helpful. Thank you for your interest in CONICS! Aaron From: dstueckm<mailto:notifications@github.com> Sent: Friday, January 24, 2020 11:10 AM To: diazlab/CONICS<mailto:CONICS@noreply.github.com> Cc: Subscribed<mailto:subscribed@noreply.github.com> Subject: Re: [diazlab/CONICS] error in classify tumor cell and nontumor cell? (#12) Hi @biobug16<https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_biobug16&d=DwMCaQ&c=iORugZls2LlYyCAZRB3XLg&r=PU3tGlVULnXYXO052fG3nSJYdab_cKD3C5PdKOVdaHY&m=huD8Vm-MjuhlhrlNjhXESnK60SX3ijsL5Aji9BdKRR8&s=IQYANviWIogbZ3XWQ9T8HYwufN0yGtUAb2wxYOZRCrY&e=>, Thanks for the reply. I should have clarified; I do understand that you can determine which clusters are normal, my point was more from an automation perspective. If I am running a large number of samples, and the histogram produced does not necessarily have any "order" in terms of which CNVs occur from left to right, it will require visual inspection each time. It would be helpful to have a function with a readout of which clusters contain which CNVs, so we could then subset the histogram in that way. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub<https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_diazlab_CONICS_issues_12-3Femail-5Fsource-3Dnotifications-26email-5Ftoken-3DAB2BEINBUWHMXZPGAEJDZ33Q7M4FRA5CNFSM4IIGQUQ2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEJ3Y5BY-23issuecomment-2D578260615&d=DwMCaQ&c=iORugZls2LlYyCAZRB3XLg&r=PU3tGlVULnXYXO052fG3nSJYdab_cKD3C5PdKOVdaHY&m=huD8Vm-MjuhlhrlNjhXESnK60SX3ijsL5Aji9BdKRR8&s=W3G3bvUTP3vQ1vG5KESFNGsJ8Fklv5IVBw6oNQCQkHw&e=>, or unsubscribe<https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_AB2BEIJI54K6GDIDLVWM2A3Q7M4FRANCNFSM4IIGQUQQ&d=DwMCaQ&c=iORugZls2LlYyCAZRB3XLg&r=PU3tGlVULnXYXO052fG3nSJYdab_cKD3C5PdKOVdaHY&m=huD8Vm-MjuhlhrlNjhXESnK60SX3ijsL5Aji9BdKRR8&s=dNO6Qt0z75ItX4XgZXeqHe4syFTOvm7i-mdb7aaWcjs&e=>.

[Ruismart]

Hi,
Thanks for your software. I want to use CONICSmat to classify tumor and nontumor cells in a 10X dataset. AND I come across two probelms.

1. Compared to tumor cell markers, I'm surprised that CONICSmat result may be reversed.
2. The color of Chromosome Heatmap is very light.
   I really appreciate your positive reponse.Thank you very much!

Best regards
Mengyan Zhu

got similar issue to test glioma datasets, and have tried different ways to modify the expression level and those cutoffs.

for 10x dataset usually it couldn't have as large nFeature/nCount as smart-seq2 data, then might use log2(CPM +1) rather than 1e4 based normalization to get a significant (brighter~) CNV pattern in the end.

about the possible automatical solution to filter for right tumor cells I think it might be tissue/tumor specific. for gliomas, could define putative 'myeloid' cell group (like microglia, T cell) using markers from the top highly variable genes 'ts' in tsne clustering plot, then after 'hi <- ...' could set '1' or '2' as normal cells according to which has the much larger proportion in these 'myeloid' cells.