Define a minimum set of terms for documenting Interactions

[zedomel]

What are the minimum set of terms which do we need to document Interactions?

Identifiers (ID's)

• One ID for the Interaction record: dwc:eventID or dwc:resourceRelationshipID
• Two ID's for related dwc:Occurrences/dwc:Taxons: resourceID and relatedResourceID
• One ID for the interaction type: dwc:relationshipOfResourceIDcommunity reviewing
• One ID for a reference: relationshipAccordingToID was proposed to be added to ResourceRelationship extension. Still needs evidence for demand add term relationshipAccordingToID to Resource Relationship extension dwc#187.

Non-identifiers

Which terms besides those already defined by DwC do we need to document Interactions?

• Interaction evidence type (e.g. pollen collect from the bee body, stomach content, direct observation)?

[zedomel]

One possible model that we have discussed on Plant-Pollinator Interactions is the Event Core + ResourceRelationship Extension:

In the diagram the dwc:Event class is used to represent an action that occurs at some location during some time, and this action being an interaction. The events may include, but not limited, spatial (dwc:Location), temporal (dwc:eventDate, dwc:eventTime) and sampling data (dwc:samplingProtocol).

The dwc:Occurrence class documents an existence of an Organism [...] at a particular place at a particular time associated to a particular dwc:Event. So, the dwc:Event is a documentation of an co-action (aka interaction) of two particular dwc:Occurrences (see Lidicker, 1979 for a discussion of co-actions and interactions in Biology).

Since, any action (or co-action) may have a direction and can be classified into different categories (or types), the dwc:ResourceRelatinship class is used to document the type of an interaction (dwc:relantionshipOfResource and the new (?) term dwc:relatinshipOfResourceID), and the direction of an interaction (dwc:resourceID --> dwc:relatedResourceID). And it is the actual representation of an interaction, the dwc:Event only represents the event the originates that interaction (basically spatial and temporal information).

Using a dwc:Event class to represent the interactions (without a direction or a type) allows generic species associations (e.g. co-ocurrences) to be documented. Which may be useful in cases where we don't have any information about the type and the direction an interaction, then we assumed dwc:Occurrence<-- interacts with --> dwc:Occurrence at some location and time (dwc:Event). But, may it does not make sense for co-occurrences (?). The dwc:ResourceRelantionship class enforces a direction to the interaction (dwc:resourceID (subject) --> dwc:relatedResourceID (object)) and also the type (dwc:relationshipOfResource (verb)) establish the direction even with generic values (e.g. interacts with).

The dwc:Event may be associated with dwc:MeasurementOrFact class to document many measurements or facts of any particular dwc:Event. Since, semantically, the dwc:Event does not represents the interaction, but rather a co-action at some place and time, it can be associated with any measurement or fact that originates to or from the interaction (e.g. resource collect from a flower, --more example here--).
Similarly, the [obis:ExtendedMeasurementOrFact(https://tools.gbif.org/dwca-validator/extension.do?id=http://rs.iobis.org/obis/terms/ExtendedMeasurementOrFact) can be used to document measurement or facts of any particular dwc:Occurrence.

It is not clear to me, if the measurement or facts should be linked to the dwc:Event or dwc:ResourceRelationship, but does not seem a problem to link them to the dwc:Event since all measurements or facts about an interaction are also, by definition, the measurements or facts of the associated event (any example that contradicts it???).

At the Plant-Pollinator Interactions we have discussed different models. See issue #60 for a discussion about advantages/disadvantages of each model.

But, the Event Core + ResourceRelationship Extension looks like the one that satisfies all the requirements to capture an Interaction:

• Organisms or groups of organisms involved in the interaction.
• The nature or classification of an interaction (interaction type).
• The direction of an interaction (interaction type + dwc:resourceID + dwc:relatedResourceID).
• An evidence for the interaction or the "interaction voucher": bibliography references, media, URI for a collection or an item in collection (e.g. pollen grains collected from a bee body, e-DNA material)

Curious to know thoughts of others!

thanks.

[tucotuco]

I feel this is a good, relatively simple yet powerful and flexible way to model interactions.

The ExtendedMeasurementsOrFacts Extension would be required to link to the Event (and Occurrences in an Occurrence Extension to the Event Core), as there is currently no possible model with ResourceRelationship as the Core in a Darwin Core archive.

[zedomel]

Hi everyone,

after our last meeting I was thinking about how we can capture the co-occurrence and the interaction separately. Since, every interaction is a co-occurrence, then it can be seen as a specialization of a co-occurrence where some kind of knowledge is added to the co-occurrence.

Based on that, for documenting co-occurrence we should need the follow ID's:

• ID for the Occurrence's (for simplicity I'm considering only two occurrences, but in fact it can be expanded to n occurring organisms or group of organisms).
  • Darwin Core terms: occurrenceID
• ID for the co-occurrence event. A co-occurrence can be documented as Event and then the eventID can be used as the ID of the co-occurrence.
  • Darwin Core terms: eventID

Once we have a co-occurrence we can add as many knowledge (ie. interaction inference) we want to that using the ResourceRelationship class of DwC. For that we have 4 (four) additional ID's:

• resourceRelationshipID: the ID of the interaction or knowledge linked to a particular co-occurrence event.
• resourceID: the ID of the subject of the interaction. It should be the ID of the Occurrence acting as the subject of the interaction.
• relatedResourceID: the ID of the object of the interaction. It should be the ID of the another Occurrence in the co-occurrence event
• relationshipOfResource (next version of DwC should include relationshipOfResourceID): the type of the interaction. Next version of DwC should include relationshipOfResourceID. Here we may use some ontology like Relation Ontology.

Additionally, we may need a term for the interaction evidence and the term relationshipAccordingTo and also the proposed term relationshipAccordingToID (in process of need evidence for demand). So, as Interest Group we can prepare an evidence for demand for the term relationshipAccordingToID.

Using the ResourceRelationship class we can also have additional information about:

• relationshipEstablishedDate: the date when the interaction inference was made
• relationshipRemarks: comments about the interaction

Thus, one co-occurrence Event can be linked to 0, 1 or many interaction inferences (ResourceRelationship):

Probably, there are some other ways to document that (e.g. linking ResourceRelationship direct to one of the Occurrences without the co-occurrence Event):

Let's discuss more about it.

[beckettws]

Hi all,
First a quick thanks for taking this on and the discussion so far. It's been super helpful for me already. I'm working with Nate Upham at Arizona State University, Jorrit Poelen, and other collaborators to extract and model mammal-virus interaction data. Separating co-occurrence of a virus in a mammal from a directed parasitic relationship is very helpful, because some viruses may be present only transiently in a mammal organism without replicating or producing disease.

I'll raise a couple further questions in case they're helpful for discussion:

1. Many published articles we're looking at report complex measurement procedures on occurrence-level interaction events but do not have vouchers and often don't even provide record-level tables. For example, the article text might describe detecting viruses in 50 bats based on non-specific antibody tests, sampling lung tissue from 5 of those 50, and finding 3 out of 5 were positive for a particular virus species based on a PCR test. I'd like to be able to explicitly describe each occurrence-level interaction record and represent the nested sampling procedures even when the paper doesn't provide a supplementary table with specimens or individual observations. (Subudhi et al. 2017) provides a more extended real-world example ("A persistently infecting coronavirus in hibernating Myotis lucifugus, the North American little brown bat"; DOI 10.1099/jgv.0.000898).
2. I'm also puzzling over how to handle experimental studies that provide evidence that a virus species infects and is transmitted by organisms in a host species. It seems analogous to the cultivated/wild distinction for species observations, but lab studies that inject a virus into a host organism or tissue sample to test for virus replication can provide some of the strongest evidence for the species having a host-parasite relationship more generally. For example, see (Griffin et al. 2021; "SARS-CoV-2 infection and transmission in the North American deer mouse"; https://www.nature.com/articles/s41467-021-23848-9).

Thanks and looking forward to more discussion soon!
Beckett