The abstract is read at the publisher; the summary is JClub's.
In an enriched biomedical knowledge graph, Inverse Node Frequency (INF) normalization improves target–adverse event association prediction for low-degree targets, boosting true positive rate (TPR/recall) versus untransformed metapath features and a standard method.
- Why it matters: Degree-driven “topological” bias can let hub nodes dominate learned signals, hurting safety inference for novel or understudied therapeutic protein targets.
- What they did: The authors apply INF to metapath-derived graph features and train/evaluate models using curated target–adverse event associations, then test on an external association set held out from modelling.
- The result: INF significantly increases recall for low-degree targets and maintains highly competitive hits@k, while also showing superior resilience on the external, previously unused target–adverse event set compared with two benchmark normalizations.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.