P values calculation not documented

[afermg]

Despite being the recommended way to compare phenotypic activity/consistensy, the usage of the p-value functions is not documented. I think it is worth adding that to the wiki/docs. Let me know where you would like to have it if I figure it out before it is up there, so I can open a PR.

[johnarevalo]

Thanks for highlighting this! p-values described in the mAP_demo.ipynb notebook are the current preferred way. The p-values are automatically computed when the average_precision scores are aggregated (e.g. per perturbation) using copairs.map.mean_average_precision

The functions shown in your search are the "deprecated" way of calculating p-values. It is obtained by computing a p_value per sample, and then aggregate. Because it has two aggregations, it would require a two-step FDR correction as well.

We should discuss whether we should still support this, or drop it. @alxndrkalinin and @shntnu may have some opinions on this.