Use embeddings to assemble link prediction classifier for human vs mouse phenotype equivalency.

[caufieldjh]

• find silver standard mouse <-> human phenotypes - possibly here Nico says here
• make silver/gold standard edge file with mouse <-> human phenotypes (or does this already exist?)

Classifiers to try:

• random forest
• MLP
• logistic regression (baseline) #11

In practice, all classifiers may be defined in a single NEAT config after implementing #5

[justaddcoffee]

@matentzn we need some mouse <-> human phenotype mappings for this OntoML project, and found this directory. Can you advise about whether we could use upheno_mapping_all.csv for this experiment?

Basically what we need is a gold or silver standard of mouse - human phenotypes in order to train a link prediction classifier. I know you probably explained what each of these files is before and which to use, just making sure we use the right file. Thanks!

[matentzn]

Great you are getting to that :)

Upheno at data.mi is outdated, use instead:

https://bbop-ontologies.s3.amazonaws.com/upheno/current/upheno-release/all/upheno_mapping_all.csv

Feel free to also plunder:

https://github.com/mapping-commons/mh_mapping_initiative/tree/master/mappings

Generally, whatever infrastructure you are building should prepare itself for reading mapping files in SSSOM format, like

https://github.com/mapping-commons/mh_mapping_initiative/blob/master/mappings/mp_hp_impc_eye_embedded.sssom.tsv

[justaddcoffee]

Great, thanks Nico!

[caufieldjh]

The gold standard phenotypes (as edges, filtered to just MP and HP) are now included as another transform, so they'll be on KG-Hub shortly.

[matentzn]

Whenever you output any mappings from any process, it would be great if you could consider publishing them in SSSOM, so we can process them further :)

[caufieldjh]

Absolutely, @matentzn - I decided to split off SSSOM generation/parsing into its own task here, but we'll definitely make them available.

[matentzn]

Sweet, because we are close to realising the sweet boomer dream, and your mappings will be very valuable, especially if they take into account graph structure!