empirical_feature_p_values that are 0

[szarnick]

Hello,

I've applied the limix_qtl processing and multiple testing corrections pipeline to our dataset multiple times now and each time I get a set of SNPs with an "empirical_feature_p_value" that is 0. How do I interpret this? Am I doing something wrong? The original p values for these SNPs are not always the highest in the data set.

[Bonder-MJ]

Hi,

Thanks for your question. This does happen. Just to check did you check the alpha_param & beta_param 's for the 0 effects? Also what are the zscores based on the beta + beta_se? These two things might help you see what is going on.

Also I'm working on an update of the running making it more memory efficient. In the update I also place the minimum P value to at the machine precision number rather than 0, this is one of the reasons why we get zero's.

Best,
Marc

[szarnick]

Thanks so much for your response! All tests with empirical p value = 0 have an alpha_param of 9.007199e+15 and a beta_param of 2. As an example, one test had a beta of 3.4974067 and a beta_se of 0.3468677.

[Bonder-MJ]

Hi, it seems to have to do with the permutations getting very extereme values relative to your actual results.
What you can do is output the permutations and have a look. Also it would be good to look at the distribution of the features/genes you are trying to test.