Import DrugBank-DrugCentral mappings

[cthoyt]

This PR imports the 3,960 mappings between molecules in DrugBank and DrugCentral that were predicted through exact string matches, manually reviewed by @caufieldjh, and stored in http://kg-hub-public-data.s3.amazonaws.com/frozen_incoming_data/drug-id-maps-0.2.sssom.tsv. Some notes:

1. DrugCentral itself only provides CAS mappings in addition to structures as SMILES/InChI.
2. This file contains novel mappings between DrugCentral and several other controlled vocabularies (ChEBI, ChEMBL Compound, Therapeutic Target Database Drugs, PharmGKB). This PR starts with DrugBank as a proof of concept. DrugBank does not contain any primary mappings to DrugCentral, as far as I can tell. The import script can be trivially updated to import some/the rest.
3. Some provenance information about what files were used to generate these mappings didn't seem strictly necessary and are not propagated through to the Biomappings file.
4. I used pyobo to add in missing labels

---

Update this PR now filters out drugcentral-drugbank mappings that are already available by querying DrugCentral's postgres database

[caufieldjh]

Thanks @cthoyt !
Let me know if/when the ingests hit any snags so I can fix the table

[bgyori]

Hi @cthoyt and @caufieldjh, thanks for working on this! Generally, Biomappings only includes mappings that aren't provided by any of the primary sources. I spot checked 10 entries from the new additions to mappings.tsv and in each case, I found that DrugCentral already lists the given DrugBank ID on the drug's landing page. For instance, taking the first new entry:

drugbank DB00001 Lepirudin skos:exactMatch drugcentral 2995 lepirudin

https://drugcentral.org/drugcard/2995 provides:

Ideally, these existing mappings would be filtered out and only novel/missing mappings added.

[bgyori]

The above is definitely on the DrugCentral website, it's odd that it didn't get filtered out - same for a couple of others that I randomly checked below.

[cthoyt]

I also ran into this for most (if not all?) that I checked. I am not sure what's going on/why they aren't in the database's xrefs table but they appear in the site

[bgyori]

The above two rows could be problematic: here, we assign a skos:exactMatch to two separate entries - these mappings that are not one-to-one could either be removed or use a different relation type.

[cthoyt]

Next step will be to identify one-many and many-one mappings and flag those for more curation (or just filter)

[caufieldjh]

Yep, those should probably be skos:narrowMatch

[caufieldjh]

For these sources, here's the relationships for mapping availability:

Source/target	CHEBI	CHEMBL	DrugBank	DrugCentral	PharmGKB.drug	ttd.drug
CHEBI to		Yes	No	No	No	No
CHEMB to	Yes		Yes	Yes	Yes	No
DrugBank to	Yes	Yes		No	Yes	Yes
DrugCentral to	Yes	Yes	Yes		No	No
PharmGKB.drug to	Yes	No	Yes	No		Yes
ttd.drug to	Yes	No	No	No	No

where "Yes" means at least some mappings are available.
It doesn't take completeness into account or that many of these are one-to-many or many-to-one since sources vary in size and specificity.