DNAcopy plateaus plot

Question

[2nelly]

Hi all,

Could someone briefly explain me why a plateau plot in DNAcopy R package can be used to define thresholds in CNA analysis?

DNAcopy manual page 4-5

Thank you in advance

Answer 1

[Noushin N]

The idea here is to come up with a threshold of logratio values, above and below which you call copy number gains and losses. Assuming that there are a few distinct copy number levels in the sample, one should see a handful of values along the y-axis.

In a sample where the majority of genome is copy neutral (CN = 2), a large stretch along x-axis will have logratio close to one. Then depending on presence of gains and losses, you should see a few other levels on the y-axis below and above 0.

The goal here is to pick a threshold value that will allow you to capture gains and losses (true positive calls in a ML/learning language), without calling copy number aberrations by mistake (false positives).

[2nelly]

Thanks for your answer Noushin! I am aware of all these you mention above. Maybe I didn t express myself correctly, so let me rephrase it: My question is how accurate can be that and additionally what is happening in cases that the plateaus are not clearly defined? I understand that for instance if you analyze some carcinoma samples the plateaus should be clearly defined due to high number of CNAs. But in other cases, like non aggressive tumors these plateaus are not so clear, right?

[Noushin N]

[Sorry about my belated reply.] Here are my guesses as to why you would get unclear plateaus:

1) Low tumor purity. You are probably aware of this fact. Low purity results in copy number levels (in LRR space) to be located too closely to each other, and thus making it difficult to come up with a threshold for calling CNAs.

2) Poor quality of array data. This will restrict the performance of your copy number segmentations. Thus, you may end up with segments whose means are some intermediate values between your expected levels, simply due to averaging values over missed breakpoints.

Since you are mentioning 'non-agressive tumors', my guess is that the tumor under discussion is smaller than what you see otherwise. Assuming that the extraction does not incorporate laser capture microdissection, you may be dealing with low tumor purity, and therefore issue (1) above.

Is this close to what you had in mind?