It is better to test only a smaller number of phenotypic features to decrease the false discovery rate and mitigate the impact of multiple testing correction. Doing less tests is better for statistics and for the environment!
I think there are 2 things that need to be done here.
First, we need to present counts of phenotypic features. We have a function that does that:
cohort: Cohort = ...
cohort.list_all_phenotypes()
This is good basic functionality, and we can add more convenience if we add term labels and (maybe) even return as a pandas DataFrame:
label
term_id
count
Seizure
HP:0001250
10
Hepatosplenomegaly
HP:0001433
4
Arachnodactyly
HP:0001166
3
...
...
...
The frame could possibly also break down the count` to the number of direct and indirect (implied by the annotation propagation rule) annotations.
Second, we need to add filter the phenotypic features prior running analysis. We already do one such filtering using min_perc_patients_w_hpo configuration option. We may want to add another filter that lets the user choose a set of HPO terms to test. Note, the ancestors of these terms will not be tested!
ielis
commented
4 months ago
It is better to test only a smaller number of phenotypic features to decrease the false discovery rate and mitigate the impact of multiple testing correction. Doing less tests is better for statistics and for the environment!
I think there are 2 things that need to be done here. First, we need to present counts of phenotypic features. We have a function that does that:
This is good basic functionality, and we can add more convenience if we add term labels and (maybe) even return as a pandas
DataFrame:The frame could possibly also break down the count` to the number of direct and indirect (implied by the annotation propagation rule) annotations.
Second, we need to add filter the phenotypic features prior running analysis. We already do one such filtering using
min_perc_patients_w_hpoconfiguration option. We may want to add another filter that lets the user choose a set of HPO terms to test. Note, the ancestors of these terms will not be tested!We need to think how to do this.
Related to #44 , #98