Refining tumor/normal assignment

[nandobonf]

Dear authors,
I successfully ran Numbat on multiple samples, however in some cases I would need to refine the tumor/normal definition (or eventually remove "noisy" CNV calls such as the BAMP). In the case shown below, for example, it is clear that genotypes 1 and 2 demote normal cells, instead, Numbat only calls genotype 1 as "normal". You can see the Numbat tumor/normal annotation labeled as "type" in the heatmap below.
Any suggestions would be highly appreciated.
Thanks in advance.

[teng-gao]

Hi,

In this case you can add up probabilities of clone_1 and clone_2 (fromclone_post) and use that as normal probability. This works because clone_post contains clonal assignment probabilities, and for each cell they sum to one.

Best,
Teng

[nandobonf]

Dear @teng-gao I am not sure I understand. Could you please provide some example code?
Thanks in advance.

[idkaavan]

Hello @nandobonf hope your project is done successfully!
It seems like you were getting a nice subclonal detection in this analysis. I was wondering if you had the ground truth so you could benchmark these results? Also, it would be truly appreciated if you could share the settings and other possible tips (e.g., input cell population or reference for healthy) that you used for your Numbat run. Thanks in advance.

[nandobonf]

Hi, the project is just started, we are facing several issues with the CNV calling because I have many tumor samples with different "normal" cell percentage (usually low) and this significantly affects the pipeline. In general, I use fibroblasts and lymphocytes as healthy reference.

[teng-gao]

Dear @teng-gao I am not sure I understand. Could you please provide some example code? Thanks in advance.

For example, if you think both clone 1 and clone 2 are normal you can just add them up for each cell like this:

nb = readRDS(url('http://pklab.org/teng/data/nb_TNBC1.rds'))
nb$clone_post %>% mutate(p_normal = p_1 + p_2)

If you don't want to curate clones through manual inspection, an alternative method is to use nb$cutree(n_cut = 1) to force number of clones to be 2, then clone 1 will be normal cells and clone 2 will be all tumor cells. Then you can use p_cnv column from clone_post as the tumor probability. Pls see this section of the manual.