2023-06-15

This is PMA's DBGI daily open-notebook.

Today is 2023.06.15

Todo today

Have a look at the DBGI discussion forum

- https://github.com/orgs/digital-botanical-gardens-initiative/discussions

Doing

Meeting on the EMI semantic model

Ana, Tarcisio and Luca

Their is a graphical schema (based on Arnaud's scheme for ENPKG here) https://app.diagrams.net/#G1g1nAgeIZEwBL63Adcwb1dDVOOO2rE2H1

https://github.com/digital-botanical-gardens-initiative/earth_metabolome_ontology

Decided to establish the licence of the semantic model to CC0
Add ORCID as contributors
protégé https://protege.stanford.edu/
Tarcisio reserved prefixes dbgi / emi at prefix.cc and purl.

Luca mentionned that we should think beforehand on the problems that could arise from having highly connected nodes (stars) such as a node ISDB linked to from all annotations with a has_ISDB_annotation edge.

We need to think about a metamodel. https://github.com/biolink

We need to think about constraints (e.g. IK should have and associated IK2D)

Ended up taking the rest of the notes at : ether pad

https://mypads2.framapad.org/p/notes-june-2023-1o1hws9v1

https://microbiomedata.org/

Hard copy of the meetings notes

Meeting on the EMI semantic model

Meeting notes 15.06.2023

Their is a graphical schema (based on Arnaud's scheme for ENPKG here) https://app.diagrams.net/#G1g1nAgeIZEwBL63Adcwb1dDVOOO2rE2H1

https://github.com/digital-botanical-gardens-initiative/earth_metabolome_ontology

Partecipants

Tarcisio Mendes
Ana-Claudia Sima
Luca Cappelletti
Pierre-Marie Allard

Comments:

- Define Meta-Model in addition to Semantic Model
- Allows topological embedding of the classes, defining fully a distance between the classes that can be used in general ML tasks
- allows explicitly defining Metadata classes (e.g. "Information Source") and properties

allows Quality Control through e.g. disjointness / other constraints, e.g. whenever InChiKey is present in the data -> InChiKey2D should be present
Can be concretely implemented as a suite of tests to run as part of GitHub repository associated to data / data quality continuous integration
Topological checks may be integrated as part of the continuous integration - e.g. adding elements that become singletons should not be allowed for release versions
Draw inspiration from BioLink Model? (https://github.com/biolink)
BioLink can be an inspiration, but has also known problem and some aspects should not be imitated
Next meeting we can list specifically these aspects which we should take care of not reproducing.
Add explicit, "verbose" descriptions to classes in addition to simple labels
Allows to use the textual embedding as part of features for ML models
how about retrieving these via cross-references whenever possible? e.g. the cross-refs to WikiData items
Logo associated to project
Possible logos may be created using tools such as MidJourney (examples provided on the Zulip platform)

For next meeting:

- Microbiological data: https://microbiomedata.org/
- Where will the data release be available from? Zenodo? Internet Archive? In what formats?
- For now ENPKG related graphs are versionned at Zenodo (https://zenodo.org/communities/enpkg/?page=1&size=20) still grouped by batches (here 96 well plates). We will work on moving to 1 profile = 1 Zenodo repo. (Note that frictionless can do quality control on Zenodo archives, would work very well on case we have the graph in a tsv format)  
- The GRAPE library can ingest TSV / CSV formats from urls, any server accessible with get requests will do
- discuss Quality Control options - at what level should constraints be defined? 
- OWL language? SHACL? Meta-model constraints + Github tests where adding bogus new elements should not be allowable