Testing Retriever

[edeutsch]

A quick way to test Retriever is to set ARAX to using TRAPI 1.6.0 KPs, since it is the only one, see e.g.: https://arax.ncats.io/?smartapi=1

A one character change in this line: https://github.com/RTXteam/RTX/blob/67da512f26661a8d16bc90c969004a2e11f2e0e5/code/ARAX/ARAXQuery/Expand/kp_info_cacher.py#L52 will set ARAX to use TRAPI 1.6.0 KPs.

I now have set up /beta to use TRAPI 1.6.0 KPs.

Using Example Query 1, the /beta endpoint now returns 0 results: https://arax.ncats.io/beta/?r=418420

Our main arax.ncats.io site is running on TRAPI 1.5.0 KPs: https://arax.ncats.io/?r=418419

Since I believe we were told that Retriever has RTX-KG2 loaded, I would have expected 41 edges from Retriever. But apparently not.

[edeutsch]

Things learned:

1. Retriever will not do any normalization. So if you send it a non-normalized CURIE, no results. Our Example Query 1 uses the CURIE CHEMBL.COMPOUND:CHEMBL112. While most other services will translate this to the correct normalized CURIE (CHEBI:46195), Retriever will not

2. If you ask PloverDB for the CHEBI:46195, 41 edges are returned and ARAX makes 22 results: https://arax.ncats.io/?r=418466

3. If you ask Retriever (via /beta) for the for the CHEBI:46195, only 24 edges are returned and ARAX makes 21 results: https://arax.ncats.io/beta/?r=418460

4. I am guessing that the factor-of-two difference in edges is PloverDB returning edges in both directions. While Retriever only returns edges in 1 direction? A guess.

5. But even if that is so, there is still a discrepancy. I wonder if there could be a version difference, too.

[edeutsch]

okay, I have restored /beta to track master and TRAPI 1.5.0 KPs again.

There is now a new arax.ncats.io/shepherd endpoint that only sends to TRAPI 1.6 endpoints (which is only Retriever)

[edeutsch]

Retriever no longer hosts RTX-KG2, but rather now it hosts DINGO data, so the Retriever vs. Plover results comparison test is no longer possible. Closing.