Beta statistical significance testing

[ielis]

• start at the "terminal" phenotypic features - HPO terms that have no descendants in the total set of phenotypic features targeted to test

• For each node

• check if we have enough counts:

  • we want to have HPO yes >2
  • we want to have variant cats >1
  • we want the total count of patients be above certain threshold
  • @pnrobinson will write a script to precalculate the p values to decide if it makes sense to perform Fisher exact test on the counts.

• if the node has "super significance" it may not be necessary to test ancestors.

  • if we know that there is k terms and the significance of this node is <p/k, this will survive even Bonferroni, so there is probably no point to test the ancestors of this node. The ancestor should be masked and not tested even for the relatives.

• look at the parents, check if the counts have changed to some extent. However, this applies only to a parent, not to all the ancestors

• marginal sum (total number of items in a row/column)

[lnrekerle]

@pnrobinson @ielis Is this also covered in the current Multiple Corrections Heuristic Strategy?

[ielis]

Yes, this issue is basically the template/draft for the heuristic MTC strategy, hence a duplicate.

We will follow up on this topic in #126