These comments aren't as applicable to collapse_qualifiers as they were for the redundant graph. For the redundant graph we needed to be able to look up the ancestors of the value of the qualifier, but to do that you need to know the qualifier value enum_name, and I wasn't sure how to derive that given an edge from ORION, so we hard coded what we needed at the time (which is still not ideal). In this case, we only care about qualifiers and values which are actually on the edge, so we should be able to capture them all. We can do this dynamically easily using the biolink model toolkit.

Getting the full list of qualifiers from the constants file was a bad idea anyway, they shouldn't be hard coded if the biolink model toolkit can supply the current list. I'm not sure why I said that. :)

instead of using this hard coded list, we can do something like the following (except you would want to instantiate the toolkit outside of the loop):

from Common.biolink_utils import get_biolink_model_toolkit
bmt = get_biolink_model_toolkit()
qualifiers = {key:value for key, value in edge.items() if bmt.is_qualifier(key)}

Then you could do something like the following, with a function that does your semantic transformations where applicable
for qualifier, qualifier_value in qualifiers.items():
qualifier_statement += semantic_adjustment(qualifier, qualifier_value)

Thanks for the suggestions. I've added a new commit that handles all of qualifiers. There is still some hard-coded decisions involved with the current qualifiers, but it will print a warning if new qualifiers that aren't handled here are ever found in edges.jsonl files in the future.

Looping through the qualifiers once you determine them like I mentioned would be better because it's more efficient (it would only attempt to handle the qualifiers that are actually there instead of checking if every possible qualifier is on every edge), and cleaner/more maintainable, mostly because the current implementation violates the DRY principle.

It would also allow you check against less things per qualifier using if/elif and remove the counting part:
if qualifier_key == qualifier_type_x:
...
elif qualifier_key == qualifier_type_y:
...
else:
print(qualifier_key not supported)

All that being said, it looks like it should all work and won't break or affect other parts of ORION, so it's fine with me to merge in if you'd like.

It would be even better to do a lookup of functions or something like that:

semantic_adjustments = {
qualifier_key_1: some_semantic_adjustment_function,
qualifier_key_2: some_other_semantic_adjustment_function,
}

if qualifier_key in semantic_adjustments:
adjusted_qualifier = semantic_adjustments[qualifier_key](qualifier_key, qualifier_value)