SciClone: Tumor purity estimation

Question

[Xiaolong Zhang]

I am trying to find a mutation-based tumor purity estimation method to due with my WES data. I was noticed that SciClone could cluster mutations of muti-region sampled tumors (remove SNVs located in CNV regions) as well as estimate tumor purities.

However, I found that the purity estimating function is closed ("doPurityScaling = FALSE") by anthor in the script of "sciClone" function. Anyhow, I edited this script to turn on this function and estimated the tumor purities of different samples. Surprisingly, the purity scores are highly unstable (range from 0.0193 to 1.6606) and many purity values were more than 1.

How dose it happens? If the purity estimating function is unusable in SciClone? If not, do you have any good method to estimate tumor purity with mutations VAF (not based on CNVs)? Thank you!

Answer 1

[Chris Miller]

As you've found, there was originally some experimental (and fairly naive) purity estimation written into sciClone, but it did not work as well as I'd like. The only thing it's doing is taking the mean VAF of the founding clone (generally cluster 1) and doubling it.

In cases where your clusters aren't well constructed (due to a small number of points, or due to CN-altered points that were not removed), that purity estimation can be pretty far off.