Get p-value for each feature-variant test

Question

Get p-value for each feature-variant test

liangyy opened this issue 5 years ago · 7 comments

Hi Natsuhiko,

I have a question about obtaining the raw p-value of the test. I am not sure if the q-value reported is always done by permutation. I am wondering if it is possible to get the raw p-value of each test (and presumably shut down permutation since it is not necessary for this setting).

Thank you very much!

Yanyu

Answer 1 · 2019-05-05T14:14:52.000Z

Hi Yanyu,

The Q-value is based on the Benjamini-Hochberg method, but not permutation. This is adjusting multiple testing for each feature.

Best regards,
Natsuhiko

Answer 2 · 2019-05-05T16:22:52.000Z

Hi Natsuhiko,

Thanks for your explanation! I am wondering if rasqual can output raw p-value w/o transforming it to q-value. If not, can I do something like p.val = q.val * rank(q.val, ties.method = 'max') / length(q.val) to get a rough estimate on p-value (I guess it is upper bound of p-value)?

Thanks!

Yanyu

Answer 3 · 2019-05-05T21:53:30.000Z

Hi Yanyu,

You can calculate the p-value from the chi-square statistic on the 11th column of the output, e.g., pchisq(col11, 1, lower=F) on R.

Best regards,
Natsuhiko

Answer 4 · 2019-05-06T01:45:21.000Z

Hi Natsuhiko,

Thanks for the clarification! I will close the issue.

Yanyu

Answer 5 · 2019-05-30T13:03:40.000Z

A followup on this, when pchisq(col11,1,lower=F) is used to get p-values, do these p-values need to be multiplied by 2 before multiple testing correction?

Answer 6 · 2019-05-30T13:10:10.000Z

You don't need to do so, the log likelihood ratio is already multiplied by 2 and this is one tailed test.

Best regards,
Natsuhiko

Answer 7 · 2019-05-30T13:12:46.000Z

Thanks for your helpful (and quick) reply