Question about relation definitions in BioHopR.
Hello,
We are using BioHopR with PrimeKG to study multi-hop biomedical reasoning and would like to reproduce your dataset construction.
The paper specifies the 12 two-hop question types as entity-type chains (e.g., Drug → Protein → Disease), but we could not determine how BioHopR question types map to PrimeKG relations. Could you please clarify the following?
Relations and traversal directions: Which exact PrimeKG relations and traversal directions were used for each of the 12 question types? More generally, how were BioHopR question types mapped to PrimeKG relations? When terms such as “associated with” or “related to” are used, do they correspond to a specific relation or set of relations between the entity types?
Treatment relations: Does “can treat” include off-label uses, or only explicit indication/treatment relations? Were contraindications and negative/absent phenotype relations excluded from BioHopR?
Gold answers: Is each gold answer list generated from the specific bridge entity in that question instance, or from all valid bridge entities in PrimeKG that satisfy the question?
Entity mapping: How were entity names mapped to PrimeKG nodes, given that names are not always unique and some entities have suffixes such as “(disease)” or unusual spacing?
For example, for the question, “Name a disease that is related to an effect/phenotype that is associated with drug Benzyl benzoate,” we would like to know which PrimeKG relations were permitted for the Drug → Phenotype and Phenotype → Disease hops.
If possible, could you also share the code or scripts used to construct the dataset? We would be happy to cite them and share our reproduction results if useful.
Thank you for your time and for releasing BioHopR. We greatly appreciate your work.
As we discussed via email, I will close the discussion here.