search

wguo-research / scCancer

A package for automated processing of single cell RNA-seq data in cancer
92 stars 39 forks source link

Description of the reference cells for inferring CNV (cnvRefData-immune.RDS) #3

Closed Puriney closed 4 years ago

Puriney commented 4 years ago

I found the reference cells for running inferCNV is a fixed file internally prepared by scCancer: cnvRefData-immune.RDS

https://github.com/wguo-research/scCancer/blob/c6078eacc2bf1ad958886c39b15e50fee92df7e6/R/cnvFunction.R#L30-L34

I have two questions:

  1. Could you describe how do you derive this file? I cannot any information in your manuscript.
  2. How robust is it to test for different cell contexts? For example, could AML and kidney cancer use the same reference?

Many thanks.

wguo-research commented 4 years ago

Thanks for your attention. The default CNV reference data in scCancer is drived from some normal immune samples, which contains T cells, B cells, NK cells, etc. By testing on some others samples(such as KC, HCC, ...), we found it is robust to estimate malignancy type generally. Currently, our package is mainly considering and testing on solid tumors, so some analysis steps may not be very applicable to AML. But for kidney cancer, I think it can be used. Besides, we have been updating the package, recently. Adding an interface for the CNV reference data is one of the new functions.

Puriney commented 4 years ago

Oh, now I see. I guess CNV/CNA inference here is non-malignant cells-centric. It is to retrieve non-malignant cells from the population so the malignant cells were indirectly identified. By contrast, the classical inferCNV is tumor cells-centric, which is to infer actual CNV events similar to the ground truth by WES. In this sense, your internal normal immune signatures make sense.