How to estimate tumor purity?

[ymc]

Typically, tumor samples contain normal cells to a certain extent. Some programs that deal with somatic mutations take tumor purity percentage as an input, e.g. ExomeCNV, VarScan, etc.

I noticed that ExomeCNV's LOH calling algorithm supposedly can be used to estimate the % contaminated by normal cells. But it is not well documented in their User Guide. Can anyone show me how to do this?

Or are there other programs that can estimate tumor purity?

Thanks in advance.

[chrisamiller]

Succinctly, you expect heterozygous somatic SNVs to be present in a sample at a frequency around 50% (and homozygous changes to be present at 100%). If your tumor is impure, these fractions will drop. If your sample is 10% normal cells, your numbers would be 45% and 90%. Regions with LOH or single copy number loss provide larger numbers of somatic mutations that should occur at nearly 100%, making this estimation easier.

[davidblaney]

Is this what you are looking for?

http://genomebiology.com/2013/14/7/R80/abstract

[ymc]

Is this what you are looking for?

http://genomebiology.com/2013/14/7/R80/abstract

looks good. Thanks a lot!

[ymc]

This is the function in ExomeCNV to estimate fraction of normal cells in tumor sample. It takes two arguments. The first one is the logR (aka log intensity) value vector for different regions indexed from 1 to length(logR). The second argument is a vector of indexes of the logR vector that is determined to be LOH.

========================================
guesstimate.contamination <-
function(logR, region.idx=NULL) {
if (is.null(region.idx)) region.idx = 1:length(logR)
med.logR = median(logR[region.idx], na.rm=TRUE)
if (med.logR < 0) { rho = 0.5 } else { rho = 1.5 }
return((2**med.logR - rho)/(1 - rho))
}
=====================================

From the look of it, it takes the median of the logR values in LOH regions and then apply this formula

let logR_m be the logR median among the LOH regions

if logR_m >=0, contamination = 3 - 2^(logR_m + 1)
if logR_m < 0, contamination = 2^(logR_m + 1) - 1

It seems to me it is theoretically possible to have contamination to be outside the bound of 0 and 1. Do you think this function is correct?

[ymc]

According to formula described in their paper, the estimation formula is

c = 1 - 2*average(|BAF_LOH - 0.5|)

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3179661/bin/supp_btr462_btr462_SupplementaryMaterials.pdf

This formula seems to make more sense. It also won't make c goes out of bound.

So which formula is correct???