question about the segment threshold

[litun-fkby]

Hi, i used the cnvkit.py segment to generate the cns file form cnr，and to call the cnv number . So i have some question about the segment.
The cnvkit version is 0.9.9
The segment command is :
cnvkit segment sample.fix.cnr -p 20s --drop-low-coverage -m hmm-tumor -o sample.segment.hmm_tumor.cns
cnvkit call sample.segment.hmm_tumor.cns -t=-1.1,-0.5,0.4,0.9 -o sample.call.hmm_tumor.cns
cnvkit scatter sample.fix.cnr -s sample.segment.hmm_tumor.cns -g ERBB2 -o sample.cnv.scatter.ERBB2.png
The png as follow:

And the log2-ratio in cns seem to be lower than the log2-ratio in cnr , and i notices that "cnvkit.py segment " have the option -t to set the threshold , so my question is:
1.The default value for the -t .
2.The ERBB2 log2-ratio in cns may be affected by the upstream and downstream gene log2-ratio, how can i set the threshold to make the segment in each gene。
And here is the cnr file and cns file :
sample.call.hmm_tumor.cns.txt
sample.fix.cnr.txt

Hope your reply.
Thanks!

[gevro]

Hello, I am also curious if anyone knows the default value for the segment command's threshold parameter?

[etal]

You can see the defaults in the help text:

cnvkit.py call -h
...
  -m {threshold,clonal,none}, --method {threshold,clonal,none}
                        Calling method. [Default: threshold]
  -t THRESHOLDS, --thresholds THRESHOLDS
                        Hard thresholds for calling each integer copy number, separated by commas. Use the '=' sign on the command line, e.g.: -t=-1,0,1 [Default: -1.1,-0.25,0.2,0.7]
  --ploidy PLOIDY       Ploidy of the sample cells. [Default: 2]
  --purity PURITY       Estimated tumor cell fraction, a.k.a. purity or cellularity.

There are some parameters in the segment command that affect how fine-grained the segmentation will be, but it's difficult to predict exactly what effect these will have; best to just try them and see.

Also consider using the -m hmm in the latest release and -m haar in the development version (source code and devel docker container). These will segment differently than the default CBS does.

As another fallback, you can use the gainloss command on the .cnr file alone to get the gene's average log2 without segmenting together with adjacent regions. For a big gene like EGFR, the gene average should be fairly reliable for spotting an amplification.