Genome music,can P-value of smg be 0?

Question

[Yuu]

Hi, I am new to Music and there was a question bothered me.

I use music to find significant mutation gene，and the result shows that there are some genes' P-value are 0.How can the P-value be 0?

Here are the code and part of my output.

Code:

genome music smg \
  --gene-mr-file gene_coverage/gene_mrs \
  --output-file Music_smg.out.xls

Output:

Muts pMbp     P-value FCPT     P-value LRT  P-value CT     FDR FCPT       FDR LRT      FDR CT
248.8         0                0            0              0              0            0
72.84         0                0            0              0              0            0
109.8         1.99E-14         0            5.66E-21       1.56E-10       0            4.45E-17
16483.52      6.22E-11         9.10E-15     5.40E-15       3.67E-07       3.58E-11     3.19E-11
11.49         2.86E-10         0            1.02E-14       1.35E-06       0            4.81E-11
37.2          3.22E-09         3.61E-13     7.42E-14       1.27E-05       8.64E-10     2.92E-10
93.68         2.38E-08         7.44E-15     2.07E-13       8.03E-05       3.51E-11     6.97E-10
117.15        3.51E-08         4.20E-12     4.54E-13       0.000103511    7.62E-09     1.34E-09

Thank you very much for your help!

Answer 1

[Chris Miller]

This generally means the p-value is below the level that can be represented by the numeric storage type. You're right that it's not exactly zero, but it gets rounded to zero when sufficiently small.

[Yuu]

Thank you very much for your answer，and I still have a question,is there a cutoff that “below the level that can be represented by the numeric storage type” in Music?

[Philipp Bayer]

I may be wrong:

In MuSiC's code from http://apt.genome.wustl.edu/ubuntu/pool/main/g/genome-music/genome-music_0.4.1-1.tar.gz in the file "ClinicalCorrelation.pm" there is a comment that says:

Internally, the input data is fed into an R script which calculates a P-value representing the probability that the correlation seen between the mutations in each gene (or variant) and each phenotype trait are random. Lower P-values indicate lower randomness, or likely true correlations.

So the lowest p-value depends on your R installation.

In R, when you run the command .Machine$double.xmin you get the lowest possible floating value on your machine, which in my case is 2.225074e-308, tiny! It may vary on your machine, especially if it's still a 32 bit machine.

[Yuu]

It helps a lot, I found the similar description in Smg.pm, thanks very much!