Disussion Topic: Instanciating reaction ontologies

[StroemPhi]

context

In the second day of last year's workshop a working group discussed the instantiation of reaction ontologies (e.g. RXNO) and the associated probems (https://docs.google.com/document/d/17xbqr4l1CeN2LNBhwCPVzo4NDkdAq4wuqEipfE7Ng-Q/edit?tab=t.0#heading=h.kww8n47po8ah)

• “We need to look into the different reaction databases and see how they map to each other, we need a hackathon for that with more time.“
• “Reaxys has their effort to develop an ontology, we need to know where they're at the moment, maybe they can join NFDI”
• Use ChemDCAT-AP to describe reactions
  • needs reaction participants defined in an ontology: e.g. reactant, solvent, catalyst, “product” as defined classes that are used to instantiate and thereby assigns the roles of the reaction participants in the respective context →
  • would need us to reach out to the participants of this group, so they can elaborate more on this need

expected outcome of the workshop

• agree on a shared pattern for how to label reaction components, how to ingest data into the reaction ontologies

prerequisites issues/tasks

• get more input about this from the people of this breakout session to better scope the task

implementation issues/tasks

[ALapkin]

Happy to take part. The person in Reaxys to discuss this with is Dr Timur Madzhidov. My team had one workshop with him when he asked our opinion about some of his ideas on reaction ontology. My team are currently looking at implementing a combination of various related ontologies (ontokin, ontospecies, …) into a task supporting processing of data from automated chemical experiments into semantically enriched datasets.

[cthoyt]

I've worked on this in ontologizing Rhea (see, https://github.com/biopragmatics/pyobo/blob/main/src/pyobo/sources/rhea.py)

I'm also thinking about this in line with the Catalaix project

[JervenBolleman]

I just want to note Rhea is already an ontology (just an RDFS one).

[JervenBolleman]

I am quite interested in extending the Rhea model for more generic reactions (non balanced) only having the key reactants.
This actually more a data-modelling exercise (where the data-model should allow reasoning for classification).
Key task is allowing for the combination of chemical tools (substructure search) and reasoning to classify if a described reaction is e.g. an is_a/rdfs:subclassOf/rdf:type "LigationReaction".

[StroemPhi]

Hi, @JervenBolleman would you be willing to chair such a session? And if yes, what would be other prerequisites we'd need to clarify?
Like when you say it needs reasoning to be able to classify a reaction and looking at the notes of this group from last year, I'd expect more parallel work in RXNO to provide the axioms for that, right?

[JervenBolleman]

I would love to chair such a session. However, I am only available a few days 4-6th. RXNO might be capable of providing the axioms, I am not sure if anyone has tried, last I looked it had no such axioms (but I did not look to closely). I suspect, that in parts this will be beyond the capabilities of OWL. Polymerization mol(n+1) == mol(n) + mol(1) is the first thing I worry about. Still we can extend OWL with rule/cheminfo based work as C3PO showed.

[StroemPhi]

Great and thank you @JervenBolleman! We'll send out more details around the agenda and regstration tomorrow. Please keep using this discussion to park any links, ideas or notes that might be helpful. I'll open up a corresponding issue (#23) that is intended to coordinate the session (in terms of sub-issues and such), where I'd like to assign you to, once you accepted the invite.

[StroemPhi]

And regarding you're availability, this session is planned to take place on the 4th, which is the day focusing on reactions and process chemistry (more info comming tomorrow), so this shoud be fine.

[StroemPhi]

A dataset example from NFDI4Cat that might be helpful for the discussion on how to instanciate reaction ontologies: https://repository.nfdi4cat.org/file.xhtml?persistentId=hdl:21.11165/4cat/32x1-99x6/9&version=1.0. Just parked here for later.

[StroemPhi]

I've created issue #23 with the intention to help the session chair better structure the hacking session during the workshop.
Please keep using this discussion for brainstorming before the workshop to add any comments, thoughts, links and tasks you think should be discussed / worked on during the session.

Being able to replay to posts directly keeps things a bit more organized here.

We'll destill this then into more actionable items within #23 shortly before the workshop.

[StroemPhi]

Hi @JervenBolleman, I'm commenting on your issue comment rahter here, so its easier to reply directly. Yes, here DCAT is not of relevance. I put ChemDCAT-AP in the issue description, because in this DCAT-AP aligend application profile we defined a ChemicalReaction shape. I'd love to improve this shape in terms of the mapping to the ontologies it currently uses to rather point to ontologies more in scope, like RXNO and PROCO instead of what we currently have with SIO and NCIT etc. I think ChemDCAT-AP might be a suitable place, because there we could define the simple contraints we want to have on instances for a reaction. But maybe this is too simple.

[JervenBolleman]

I think that there a number of issues with the ChemicalReaction shape. First I don't think it can do stoichiometric coefficient at first look, also I would decompose the "Reaction Formula" from the "Context" that a Reaction occurred in. This is why I want to have the post-its so that people can model the concepts independently and then build a consensus that captures the rich divergent needs of the community.

[StroemPhi]

Reagent (which should actually be Reactant, but there is no such class yet anywhere), Catalyst and StartingMaterial, all have a slot pointing to their MolarEquivalent, but can totally skip ChemDCAT-AP in the session. I'djust love to be able to put the most important aspects we can agree on into its next version. So, post-its you shall recieve 😄

Can you maybe elaborate more on what you mean with: "I would decompose the "Reaction Formula" from the "Context" that a Reaction occurred in"? I have an idea, but would rather not speculate.

About the "Context", I'm very optimistic @ALapkin will provide much food for thought.

[JervenBolleman]

For context, lets take the idea of the "Reaction Formula" being the oxidation of hydrogen. The "Context", can be a fuel cell or a rocket engine ;) the "Context" describes catalysts, temperature, pressure, specific reaction vats at a certain date. While the "Reaction Formula" would just be 2H + 0 => H20 (sorry for the terrible notation in this example)

[StroemPhi]

OK, then my interpretation of your words would be: the "reaction formula" should be that ontology class (here "hydrogen oxidation"), which can, at best, be axiomatized in an ontology such as MOP & RXNO, and the "Context" is something that needs to be defined in other ontology modules, such as PROCO, OntoKin, ChEBI etc. The reaction module of ChemDCAT-AP could then serve as that shape profile, which encodes the absolute minimum information needed to instanciate a reaction class given its most general context. More elaborate profiles could then be build on that to allow a much finer context description.