Question: SomaticSignatures: Compare my datas with validated mutational signatures
3
5.1 years ago by
France
muller.etienne30 wrote:

Hi,

I'm currently using SomaticSignatures package to extract signatures from my NGS datas.

I wanted to compare it to validated mutational signatures published by Alexandrov et al. (ALEXANDROV, Ludmil B., NIK-ZAINAL, Serena, WEDGE, David C., et al. Signatures of mutational processes in human cancer. Nature, 2013.) , to evaluate the implication of each validated profile in my datas.

I got their profiles on their server (ftp://ftp.sanger.ac.uk/pub/cancer/AlexandrovEtAl) and included them in the matrix obtained from mutationContextMatrix function (= sca_occurence in tutorial).
Then I applied the 3 statistical methods (nmf, pca and kmeans). But the tool pools and calculates again all the mutational profiles.

Is there a method to fix validated profiles and just having an estimation of their implication in my datas?

Thanks

modified 4.4 years ago by chang02_2320 • written 5.1 years ago by muller.etienne30
Answer: SomaticSignatures: Compare my datas with validated mutational signatures
3
5.1 years ago by
Julian Gehring1.3k
Julian Gehring1.3k wrote:

The next release of the SomaticSignatures package contains the 21 mutational signatures of the Alexandrov paper as a data set. This version will be part of the upcoming Bioconductor release which will come out this month.

Currently, I am not aware of a stable out-of-the-box method for identifying a set of signatures in the observed data.  I have been working on this over the last months.  This will need further testing, and I expect to included them officially in the package soon (this fall or winter).

I've just updated to v2 along with BioC v3 and thought this is a superb package.  I would like to analyse my data using the new "21 signatures" data but could you confirm the only way of doing so is using them as described by Muller in his original question please?

Best,
Dave

The approach described in the original question is not really a way to do it at all, and you should not use for this purpose. Estimating the existing of already signatures requires a different approach. As I wrote in the answer to Etienne's question, this will be available in the future.

is the comparison available in the current release 2.4.5?

A comparison to published signatures should be fairly easy if one has defined a suitable measure for comparing the identified signatures. If we go with your example of the KLD, we can compare the matrix we get from samples(sigs) to the 21 signatures published first on this. They are included in the package and you can access them with data(signatures21).

Do you mean to compare signatures(sigs) to data(signatures21) because both will return mutation motifs (e.g. CA A.A) as rows and signatures as columns?

Yes, both matrices have the same structure and one can e.g. use a reasonable distance measure to compare the two.

Any update on this? I'd like to compare my signatures with the 30 COSMIC signatures found here:

http://cancer.sanger.ac.uk/cosmic/signatures

When I use SomaticSignatures on my dataset, I see a difference between the signatures determined with the methods of the package and the signatures from Alexandrov: the sum of each Alexandrov signature is 1 but the sum of each determined signature is between 18 and 22 (sum(as.numeric(sigs_nmf@signatures[,1]))).
It is probably not a good idea to compare such different vectors.

Is it normal to get sum beyond 1? Should I normalize the vectors by their sum?

Answer: SomaticSignatures: Compare my datas with validated mutational signatures
0
5.1 years ago by
France
muller.etienne30 wrote:

Thanks for your response. I look forward to the next version!

Answer: SomaticSignatures: Compare my datas with validated mutational signatures
0
4.4 years ago by
chang02_2320
United States
chang02_2320 wrote:

I also read in a paper (http://www.nature.com/ng/journal/v47/n7/full/ng.3335.html) by Katenin et al, in the methods section - mutation signature analysis, They compare their signature to Alexaddrov using Kullback-leibler divergence. Maybe this a viable way to compare.

From the paper

"The final signatures were extracted as in Alexandrov et al. This process yielded three mutational signatures. The signatures p obtained were compared to the published signatures q of Alexandrovet al. by mean Kullback-Leibler divergence (DKL(p||q) + DKL(q||p) ]/2."