Work

Datasets and Models for Plausibility-Driven Prioritization of Candidate Biomedical Annotations

Year: 2026
Type: preprint
Abstract: Datasets and models for reproducing the experiments presented in the article "Plausibility-Driven Prioritization of Candidate Biomedical Annotations". The deposit contains five ZIP archives, one for each biomedical knowledge graph (bioKG) considered in the study: - miRNA-KG- Hetionet- PKT-KG- PrimeKG- OptimusKG Each archive contains: - Graph data: nodes and edges in CSV format.- Negative samples: community-based negative examples generated separately for each schema fact.- ML models: relation-specific Random Forest classifiers.- Plausibility scores: scores computed using the four formulations evaluated in the article: - Base - Gain - Combo - SoftMax A schema fact identifies a relation type through its source entity type, predicate, and target entity type. For example: "miRNA - regulates activity of - Gene" Each ZIP archive follows this structure: /├── data/│ ├── _nodes.csv│ └── _edges.csv├── negative_samples/│ └── c-b-n-s/│ └── .csv├── dumps_models/│ └── transe/│ └── c-b-n-s/│ └── err_beta_1/│ └── RF/│ └── RF_model_ .pkl└── plausibility_scores/ ├── base_scores_ _c-b-n-s.pkl ├── base_scores_ _neg_c-b-n-s.pkl ├── gain_scores_ _c-b-n-s.pkl ├── gain_scores_ _neg_c-b-n-s.pkl ├── comb_scores_ _c-b-n-s.pkl ├── comb_scores_ _neg_c-b-n-s.pkl ├── softmax_scores_ _c-b-n-s.pkl └── softmax_scores_ _neg_c-b-n-s.pkl "c-b-n-s" denotes the community-based negative sampling strategy. Files containing "_neg_" report scores assigned to generated implausible triples, whereas the corresponding files without "_neg_" contain scores assigned to blind plausible triples held out before embedding generation and classifier training. Finally, the deposit includes the case studies mentioned in the article (curation_case_study.tsv and temporal_case_study.tsv).
Institution: University of Milan
Open Access status: green
OpenAlex ID: W7170157579
Cites:
Cited by:
Related to: