Drugbank: Review chemical <-> Gene/Protein predicates

[colleenXu]

Is your feature request related to a problem? Please describe. The issue is picking a predicate for drugbank descriptions of genes/proteins that are linked to chemicals (keywords like enzyme, carrier, transporter, inducer, target, etc.). This data is ingested by MyChem.info.

For example, with enzyme: "interacts with" seems too broad. "affects metabolism of", "increases metabolism of" doesn't seem right (its the actor that does the enzymatic/metabolic reaction, which seems more direct than these predicates imply).

What working group (or team) did this request originate from? MetaKG related (Service Provider, Exploring Agent)

Describe the solution you'd like

• for basic enzymatic reactions:
  • Gene/Protein -"has substrate"-> ChemicalSubstance (this can be close map to DRUGBANK:enzyme and SIO:000905 "has substrate")
  • ChemicalSubstance -"is substrate of"-> Gene/Protein (this can close map to DRUGBANK:enzyme, close/exact map to DIDEO:00000041 "is substrate of", close map to NCIT:R122 "Chemical_Or_Drug_Is_Metabolized_By_Enzyme")
• note that I can't find NCIT:enzyme_metabolizes_chemical_or_drug

Follow a similar process for other concepts in drugbank (https://dev.drugbank.com/guides/terms), to make "more direct" predicates (ex: "transports"/"transported by" rather than "affects transport of").

[sierra-moxon]

Thanks @colleenXu! - see #478 as well?

[colleenXu]

I don't think it's the same as #478 since these aren't full reactions being modeled. I imagine a complete reaction as having "products" and these don't. It is more of a direct chemical <-> gene/protein relationship.

For example: relationships like carrier/transporter are not "reactions" (more like temporary physical interactions that have biological effects/effects on where the chemical and protein are in the body/cell).

[colleenXu]

Edit: note that DRUGBANK:target has a specific definition, that is not addressed with its current mapping to "physically interacts with".

From the glossary: the drug-target binding changes the normal function of the target, leading to effects (therapeutic or adverse).

Could have following predicates in biolink instead:

• "has target": maps to SIO:000291 ("has target"), NCIT:R146 ("Chemical_Or_Drug_Affects_Gene_Product")
• "targeted by": maps to GENO:0000634 ("is_targeted_by"),

[colleenXu]

Maybe related to https://docs.google.com/spreadsheets/d/1VI2s5YVDwOHiQv7hBWkl12d86yWvm-z_pDJFnDwsoiY/edit#gid=1053404343

[ehinderer]

Is the issue that "Physically interacts with" is not specific enough for DrugBank's definition, since DrugBank requires that the drug physically interact AND alter the normal function of the molecule?

[colleenXu]

@ehinderer I believe you're talking about the "target" post above (and not the "enzyme" example in the first post)?

Yes, I do mean something along those lines. "physically interacts with" doesn't seem to capture the info for "target" vs "carrier/transporter" vs "enzyme". I think "target" and "enzyme" would at least be helpful for Translator use cases....

[cbizon]

One thing that always bothers me about 'target' is that it seems as though it's not purely a binary relationship between a chemical and a gene/protein. That is, usually we talk about something being a target in the context of a particular disease. A drug can have functional effects on other proteins as well, it's just that for whatever the ostensible purpose of that drug is, those other effects are considered "off-target". But change the endpoint, change the target.

[colleenXu]

I don't see a particular issue with having multiple targets. I wonder if other resources take an approach similar to drugbank, where a chemical can have multiple "target" gene/proteins, because a "target" is just what will interact and be affected by the chemical.

I seems that one needs a particular context (this disease/body state vs another, one desired effect vs another) in order to designate what "off-target" means.

[sierra-moxon]

DrugBank is not public domain any longer :( - @vdancik gave us this info License prohibits redistribution.

[sierra-moxon]

Talked about this issue on July 25 helpdesk with Guangrong - there seem to be two different statements we are trying to make:

1. chemical -> targets -> gene/protein (or some gene/protein in a pathway)

• drug->targets->gene/protein->associated with->disease
• drug -> targets -> gene/protein -> interacts with -> gene/protein -> associated with -> disease

2. gene/protein->is_druggable_for (targets)->disease

we probably need two predicates, or we need to agree on a strategy for either making these statements into a "one-hop" representation or not.