signature per sample

[aheravi]

Hi @dami82 ,
I need to have the most relevant signature per TCGA samples. Please comment on my take below.

# Retrieve exposures (results)
tcga.exp <- tcga.analysis$Results$exposures
tcga.sign = as.data.frame(tcga.exp)

Then considering a head like below,

>   tcga.sign[,1:3]
        TCGA-02-0003-01 TCGA-02-0033-01 TCGA-02-0047-01
Sign.01       4.7028706     10.03246751      19.2853478
Sign.02       0.6020939      0.06459751       0.4231106
Sign.03       3.8585791      1.19942985      13.8536437
Sign.04      35.3896970     17.31033310      29.7662441
Sign.05      13.5525516      8.37284467       9.1487803
Sign.06       2.8942077      2.02032736      11.5228735

I'm concluding that all the samples' signatures could be "Sign.04" for the samples shown above (as the reported values for that signature is more than the other signatures).
Is that a valid conclusion?
Also, what those numbers are, and should I consider a threshold for defining a sample signature based on that?

Thanks!

[dami82]

Hi aheravi,Thanks for using the library and for your question.1) these numbers are "reconstructed expected mutation counts". If you run a colSums, you should get aprox the input total mutation count. 2) Your interpretation is correct (in those 3 samples, sign 4 is prevalent). Finding the max for each column (sample) should suffice. Setting a threshold may be better, but it depends on your hypothesis. 3) Additional notes. Make sure that the assumptions make sense (e.g., is k=6 a reasonable number of signatures to extract?). Also, you can match these signature profiles with COSMIC if you want (either version 2 or version 3, i.e., SBSnn).Best regards,Damiano——Damiano FantiniSent from my iPhoneOn May 19, 2023, at 17:03, aheravi ***@***.***> wrote: Hi @dami82 , I need to have the most relevant signature per TCGA samples. Please comment on my take below. # Retrieve exposures (results) tcga.exp <- tcga.analysis$Results$exposures tcga.sign = as.data.frame(tcga.exp) Then considering a head like below,

tcga.sign[,1:3]

TCGA-02-0003-01 TCGA-02-0033-01 TCGA-02-0047-01 Sign.01 4.7028706 10.03246751 19.2853478 Sign.02 0.6020939 0.06459751 0.4231106 Sign.03 3.8585791 1.19942985 13.8536437 Sign.04 35.3896970 17.31033310 29.7662441 Sign.05 13.5525516 8.37284467 9.1487803 Sign.06 2.8942077 2.02032736 11.5228735 I'm concluding that all the samples' signatures could be "Sign.04" for the samples shown above (as the reported values for that signature is more than the other signatures). Is that a valid conclusion? Also, what those numbers are, and should I consider a threshold for defining a sample signature based on that? Thanks! —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you were mentioned.Message ID: ***@***.***>