RO terms for drug-target interaction

[sebotic]

Hi Chris and team,

Sebastian from Andrew Su's lab here. We have been working out a set of drug to disease relations which we plan on integrating into Wikidata. As far as I could see, such terms do not exist yet in RO. For interoperability, it would be great if you would be willing to add these terms also to RO, so we could xref directly.

The list of terms is actually what ChEMBL also uses to describe the mechanism of action of a certain compound. In total, it would be 29 terms, but for now, we would be happy if you could integrate the 10 most relevant to us.

These are:

• Activator
• Agonist
• Positive allosteric modulator
• Positive modulator
• Inhibitor
• Antagonist
• Negative allosteric modulator
• Negative modulator
• Disrupting agent (e.g. protein complexes)
• Blocker

The whole schema we have in mind for Wikidata: https://www.dropbox.com/s/l7yvvhdj2ybo94x/Drug-Target%20interaction%20data%20model.png

Thanks!

Best, Sebastian

-- Sebastian Burgstaller-Muehlbacher, PhD Research Associate Andrew Su Lab MEM-216, Department of Molecular and Experimental Medicine The Scripps Research Institute 10550 North Torrey Pines Road La Jolla, CA 92037

[cmungall]

Great! We have some already and can add synonyms. Others will be new relations. @mbrush is this something you can do as a PR?

[mbrush]

@sebotic does chembl provide documentation/definitions for these terms that we can use/reference? Many are obvious, but some are not entirely clear w.r.t. their meaning (e.g. 'blocker').

Also, most of the terms in the full list of 29 make sense as relations between a drug and its molecular target - rather than a relation between the drug and disease it treats (e.g. 'allosteric modulator', 'releasing agent', 'substrate', 'chelating agent'). Is this your intent? Or did you want to try and frame these as relations directly connecting a drug to a disease (which could get awkward)?

[mbrush]

One more thing - curious if you considered a design pattern where some or all of these properties are modeled as roles/functions inhering in the drug - rather than encoding them as relationships? I am not familiar enough with Wikidata modeling principles to know if they try to avoid proliferation of properties, or if they have design patterns for n-ary relations that could be used instead to capture things like the role a participant plays in an interaction. But if this is a viable approach, note that chemical ontologies like ChEBI already have 'role' classes representing many of the concepts you would need (e.g. 'agonist', 'antagonist', 'modulator', 'chelator'), and those missing could easily be added.

I could then imagine a couple approaches for creating ternary relations that describe the role a drug plays in its interaction with a particular target to treat a particular disease. For example, use a single 'has mechanism of action' property to create claims that link a drug to a role/mechanism, and qualify this by the particular disease or target against which they are applied where this role is realized. Or create a claim that the drug 'has molecular target' some drug, and hang the specific role/mechanism as a qualifier from this claim.

As I said, not sure if this type of thing is done in Wikidata, but wanted to put it out there as an alternative to creating 29 new properties to describe these roles.

[sebotic]

@mbrush ChEMBL, currently does not provide an ontology or official documentation/publication of these relations/interaction types. (blocker would be a substance which block e.g. an ion channel protein). We got this list through our friends at ChEMBL. It seems that there are no immediate plans to turn these into an ontology or official document.

The interaction types listed here are solely for drug - target interactions, yes. For drug - disease, we have a different set of interactions:

• drug used for treatment (to describe which drugs can be used for a disease (http://tools.wmflabs.org/sqid/#/view?id=P2176&lang=en)
  • medical condition treated (to describe which drugs may treat what condition (http://tools.wmflabs.org/sqid/#/view?id=P2175&lang=en))
• cause of (to describe side effects (http://tools.wmflabs.org/sqid/#/view?id=P1542&lang=en))
• significant drug interaction (http://tools.wmflabs.org/sqid/#/view?id=P769&lang=en)

Only thing which is still missing is a 'may prevent' edge.

Regarding using roles/functions: We have thought about that and have imported many compounds from ChEBI but not the ontology structure yet. That would be doable. A while back, I also established a 'has role' property. We already had a model in place and also imported data based on that model (https://www.dropbox.com/s/idoyyw9dllvcqhd/Drug-Target%20interaction%20data%20model_original.png?dl=0) and just says: compound x "Physically interacts with" protein y, and a qualifiers specifies what the type of interaction is (e.g Agonist, Inhibitor). But we decided that direct relations are more useful for representing that knowledge and also for using it in sparql queries.

In order to represent the info we'd like to capture, both "physically interacts with" or "has role" would require to add qualifiers, making the data model more clunky. Although it is some additional work to get the properties approved, we think that, on the long run, it should be worth it, as it allows a nice property and sub-property model.

[mbrush]

Ok. Thanks for the explanation @sebotic. I am starting to draft definitions for these relations and fit them into the RO hierarchy. One important distinction to make for a few of these terms these is whether physical binding is required between a drug and target in effecting the outcomes listed above. That is, do you want the definitions for relations like 'is activator of' and 'is positive modulator of' to specify that the effector molecule exerts its effect through physical interaction with the target molecule? This will have bearing on how I map these relations to existing terms such as RO_0002450 ('molecularly increases activity of'), which does not expressly require the effector molecule to directly interact with the target.

Generally in these cases I tend toward creating the more generic/inclusive relation (i.e. not requiring a physical interaction). But given your comments above, and the use case/source that is requesting these terms, it may be that these relations should specify a necessary physical interaction between effector (e.g. drug) and target in all cases. Obviously for many of the terms a physical interaction is implied - uncertainty exists only around Activator, Positive modulator, Inhibitor, Negative modulator. Please advise.

[mbrush]

Made some initial label and definition proposals in gdoc [here] (https://docs.google.com/document/d/1b_nrq97WWLlHCXzaAkCPyr6NAHJUz2KycKTdPVJE-OQ/edit#). Open for comment and discussion.

[mbrush]

Making a formal proposal here for these ten drug-target interaction relations - including labels, definitions, and hierarchical classification. These relations will live under RO:0002448 ('molecularly controls'), and the textual definitions proposed below are patterned after that of this 'molecularly controls' property ("Holds between molecular entities a and b when the execution of a activates or inhibits the activity of b").

Relations were requested for the following concepts:

• Activator
• Agonist
• Positive allosteric modulator
• Positive modulator
• Inhibitor
• Antagonist
• Negative allosteric modulator
• Negative modulator
• Disrupting agent (e.g. protein complexes)
• Blocker

Given that these terms were based on annotations found in Chembl, and will be used to describe mechanism of action for drugs on their molecular targets, the definitions below were written to reflect the domain-specific meaning of the requested concepts in the field of molecular pharmacology. For example, the 'Inhibitor' concept is translated into the relation 'decreases enzymatic activity of' (as opposed to being mapped to the existing RO_0002449 ! 'inhibits' relation) - because in the molecular pharmacology domain 'Inhibitor' is used specifically to describe drugs that act on enzymatic targets (vs receptors or channels). RO_0002449 will remain as a more general relation that applies to any form of molecular inhibition. See here and here for a more complete characterization of the domain-specific meaning of the requested terms, which guided the definitions and classification below.

I. Label and Definition Proposals

1. Activator Relation Label: 'increases enzymatic activity of' Definition: Holds between molecular entity x and enzyme y if and only if x executes a process that directly activates an enzymatic process performed by y.

2. Inhibitor Relation Label: 'decreases enzymatic activity of' Definition = Holds between molecular entity x and enzyme y if and only if x executes a process that directly diminishes an enzymatic process performed by y.

3. Agonist Relation Label: 'is agonist of' Definition: Holds between molecular entity x and receptor molecule y if and only if binding of x to y activates a process that realizes a biological function of y.

4. Antagonist Relation Label: 'is antagonist of' Definition: Holds between molecular entity x and receptor molecule y if and only if binding of x to y diminishes the effect of a primary agonist on receptor activity.

Question - we may want to tweak this definition if 'antagonists' can diminish the activity of constitutively active receptors that require no activating agonist.

5. Positive modulator Relation Label: 'is positive modulator of' Definition: Holds between molecular entity x and receptor molecule y if and only if x executes a process that directly enhances a process executed by y. (Alternatively: Holds between molecular entity x and receptor molecule y if and only if binding of x to y enhances the biological activity of y)

Question here for @sebotic - It is not clear to me what distinction exists, if any, between 'positive modulator' and 'agonist'. As written here, it may be that activation by an agonist requires x binding to y, which this might not be required for modulation? Or perhaps that agonists initiate target activity, while modulators may initiate or enhance existing activity? Or perhaps the distinction is not needed and we can eliminate 'positive modulator'? Perhaps a contact of yours at Chembl can review this and other definitions here?

6. Negative modulator Relation Label:'is negative modulator of' Definition: Holds between molecular entity x and receptor molecule y if and only if x executes a process that directly reduces an process executed by y. (Alternatively: Holds between molecular entity x and receptor molecule y if and only if binding of x to y diminishes the biological activity of y)

7. Positive allosteric modulator Relation Label: is positive allosteric modulator of' Definition: Holds between molecular entity x and receptor molecule y if and only if x binds to y at a regulatory site distinct from that of a primary agonist, and enhances the biological response induced by the primary agonist.

8. Negative allosteric modulator Relation Label: 'is negative allosteric modulator of' Definition: Holds between molecular entity x and receptor molecule y if and only if x binds to y at a regulatory site distinct from that of a natural ligand, and diminishes the biological response induced by the natural ligand.

9. Disrupting agent Relation Label: 'is disruptor of' Definition: Holds between molecular entity x and molecular complex y if and only if binding of x decreases the cohesiveness or stability of y.

10. Blocker Relation Label: 'is blocker of' Definition: Holds between molecular entity x and molecular channel y if and only if binding of x to y impedes transport through y.

II. Classification Proposal

• molecularly controls* (RO_0002448)
  • molecularly increases activity of* (RO_0002450)
  • increases enzymatic activity of
  • is positive modulator of
    • is agonist of
    • is positive allosteric modulator of
• molecularly decreases activity of* (RO_0002449)
  • decreases enzymatic activity of
  • is negative modulator of
  • is antagonist of
  • is negative allosteric modulator of
  • is blocker of
  • is disruptor of

(starred terms above currently exist in RO)

@sebotic, as noted above, it would be great to have a domain expert vet these term definitions and their hierarchical classification. Do you know anyone . . . maybe a Chembl contact?

[mbrush]

Just a note that we should we aware of additional drug-target interaction types as enumerated by other resources. DGIdb enumerates 33 (link) and Drugbank also has an enumerated type system that populates its 'Actions' attribute (link).

If this is the modelling approach we will use to describe this aspect of pharmaco-biology, we will need a coherent and comprehensive hierarchy of relations to which we can map all 'interaction types' from different data sources. And we need to be sure that modeling interaction type using a hierarchy of specific relationships (as opposed to a qualifier on a more general association) will accommodate all use cases.

[mbrush]

@sebotic, others. Hoping someone can review the proposal above . . . ideally someone with some domain knowledge in pharmaco-biology, but any second pair of eyes will do. Would like to get these into RO and mapped to WD soon.

[mbrush]

Found definitions for interaction types used at the DGIdb here: http://dgidb.genome.wustl.edu/getting_started (scroll down a bit for the definitions). Use these to vet the definitions proposed above.

[nlharris]

Is this still in progress?