more specific operations for MyChem chembl.drug_mechanisms data

[colleenXu]

Related to biothings/biothings_explorer#532 (comment)

MyChem chembl.drug_mechanisms data, in subject-association-object format

• We want 1 association per unique combo of subject/object/action_type.
  • Right now in MyChem, all data is aggregated by unique chemical (chem-centric format). So when a chemical has multiple drug_mechanisms, we can't retrieve only the sections that have a specific action_type.
  • MyChem currently has info for 5262 unique chemicals
• We want the gene ID to be NCBIGene (or UniProtKB or ENSEMBL ENSG maybe).
  • Right now, it's CHEMBL.TARGET and Translator's Node Normalizer doesn't have cross-mappings/name-retrieval for that ID namespace

[colleenXu]

Related issue: biothings/biothings_explorer#316, JQ development work

[colleenXu]

One reason to create a pending api / adjust the parser for that resource: the gene ID being CHEMBL.TARGET is a problem that isn't solved by post-processing (JQ).

[newgene]

Cross-ref a related issue here: biothings/mygene.info#105 (mapping from CHEMBL Target ID to gene id)

[colleenXu]

Notes on the data

(6306 records total https://www.ebi.ac.uk/chembl/g/#browse/mechanisms_of_action/)

Drugs

Not all drugs are "Small molecule". In rough order most to least:

• almost 75% are "Small molecule" (4727)
• almost 14% are "Antibody" (866)
• other:
  • Protein
  • Unknown
  • Enzyme
  • Oligonucleotide
  • Oligosaccharide
  • Gene
  • Cell

Targets

Not all targets are human stuff. In rough order from most to least:

• almost 83% are "Homo sapiens" (5229)
• others are bacteria, virus, fungi, or - N/A - targets

Mot all targets are proteins. In rough order from most to least:

• almost 69% are "SINGLE PROTEIN" (4155)
• almost 11% are "PROTEIN FAMILY" (667)
• Other:
  • - N/A -
  • PROTEIN COMPLEX
  • PROTEIN COMPLEX GROUP
  • NUCLEIC-ACID
  • SMALL MOLECULE
  • PROTEIN NUCLEIC-ACID COMPLEX
  • OLIGOSACCHARIDE
  • MACROMOLECULE
  • METAL
  • SUBCELLULAR
  • CHIMERIC PROTEIN
  • LIPID
  • SELECTIVITY GROUP
  • PROTEIN-PROTEIN INTERACTION
  • UNKNOWN
  • ORGANISM
  • CELL-LINE

categories of drug mechanisms

when browsing chembl https://www.ebi.ac.uk/chembl/g/#browse/mechanisms_of_action (roughly in order of most to least):

• INHIBITOR
• ANTAGONIST
• AGONIST
• - N/A - 🧇
• BINDING AGENT 🧇
• BLOCKER
• MODULATOR 🧇
• POSITIVE ALLOSTERIC MODULATOR
• HYDROLYTIC ENYZME
• ACTIVATOR
• Other (in alphabetic order, not most to least)
  • ALLOSTERIC ANTAGONIST
  • ANTISENSE INHIBITOR
  • CHELATING AGENT
  • CROSS-LINKING AGENT 🧇
  • DEGRADER
  • DISRUPTING AGENT
  • INVERSE AGONIST
  • NEGATIVE ALLOSTERIC MODULATOR
  • NEGATIVE MODULATOR
  • OPENER
  • OTHER 🧇
  • OXIDATIVE ENYZME
  • PARTIAL AGONIST
  • POSITIVE MODULATOR
  • PROTEOLYTIC ENZYME
  • REDUCING AGENT
  • RELEASING AGENT
  • RNAI INHIBITOR
  • SEQUESTERING AGENT
  • STABILISER
  • SUBSTRATE
  • VACCINE ANTIGEN

🧇 not as helpful for the creative-mode issue 532

[colleenXu]

Example of current MyChem structure vs association-based structure

Background

ANG1005 (CHEMBL1089636) has two drug-mechanisms that are different categories:

• BINDING AGENT for Prolow-density lipoprotein receptor-related protein 1 (chembl target CHEMBL4630884 aka Q07954)
• INHIBITOR for tubulin (chembl target CHEMBL2095182 aka a bunch of UniProt accessions because it's a protein complex group)

In MyChem

these two drug-mechanisms are nested inside the chembl.drug_mechanisms field of the MyChem record for this chemical: https://mychem.info/v1/query?q=_exists_:%22chembl.drug_mechanisms%22%20AND%20chembl.molecule_chembl_id:CHEMBL1089636. This means BTE post-processing (JQ?) is needed to retrieve only the INHIBITOR drug-mechanism (or vice versa).

{
    "chembl": {
        "molecule_chembl_id": "CHEMBL1089636",
        "drug_mechanisms": [
            {"action_type": "INHIBITOR", "references": {...}, "binding_site_name": null, "target_chembl_id": "CHEMBL2095182", "target_uniprot_accession": ["P68371", ...]},
            {"action_type": "BINDING_AGENT", "references": {...}, "binding_site_name": null, "target_chembl_id": "CHEMBL4630884", "target_uniprot_accession": "Q07954"}
        ]
    }
}

association-based structure

However, if we use an association-based structure, we can make two separate records. And these two records can be retrieved separately depending on what association.action_type is set to when querying.

{
    "subject": { "drug_chembl_id": "CHEMBL1089636", ...},
    "association": { "action_type": "INHIBITOR", "references": {...}, "binding_site_name": null},
    "object": { "target_chembl_id": "CHEMBL2095182", "target_uniprot_accession": ["P68371", ...]}
},
{
    "subject": { "drug_chembl_id": "CHEMBL1089636", ...},
    "association": { "action_type": "BINDING_AGENT", "references": {...}, "binding_site_name": null},
    "object": { "target_chembl_id": "CHEMBL4630884", "target_uniprot_accession": "Q07954"}
}

[rjawesome]

I've started a pending API python script if the association-based structure is preferred. Will post repo soon.

[colleenXu]

Note, I'm not sure how to handle the records that may lack an "action_type" value...they seem to lack a lot of information...

https://mychem.info/v1/query?q=_exists_:chembl.drug_mechanisms%20AND%20(NOT%20_exists_:%22chembl.drug_mechanisms.action_type%22)&fields=chembl

[rjawesome]

See https://github.com/rjawesome/mychem-drug-mechanisms

[colleenXu]

At the moment, creating a new API is not necessary.

• target IDs (ensembl gene or uniprotkb) were added for chembl.drug_mechanisms, can we add CHEMBL.TARGET ID mapping to UniProt? mychem.info#151
• Wrote more general operations for now to cover this dataset
  • for biolink 3.1.1 (currently on dev/CI instances): NCATS-Tangerine/translator-api-registry@db32e89
  • for biolink 2.4.8 (currently on test/prod instances): NCATS-Tangerine/translator-api-registry@83775ed
• Haven't done this yet: using MyChem's new list_filter function (discussed here Use JQ api-respone-transform.js#38 (comment)), I could write operations in the chem -> gene direction that are more specific (however not all records would be covered). See New API based on MyChem drugcentral.bioactivity data #101 (comment) for an example

[colleenXu]

Leaving open; in the future, we may want to write more specific operations (see the third bullet point "haven't done this yet" in the post above). I therefore moved this issue to "on-hold"

[colleenXu]

Closing for now: will open another issue to consider writing more specific operations using filter/jmespath