Support-aware evaluation of release sensitivity in a biomedical knowledgebase
Abstract
Versioned biomedical knowledgebases are widely reused as analytical inputs, but a release identifier alone does not reveal whether downstream rankings remain stable. This study constructed a reusable, support-aware data record that separates common-pair score revision, fixed-roster reranking, release-native roster turnover and entity-mapping evolution across Open Targets Platform releases 25.12, 26.03 and 26.06. The record contains harmonised row-level associations, mapping states, panel membership, ranking and set-transition metrics, robustness outputs, source manifests and validation contracts. Across 2,313,492 persistent disease-target pairs and 8,491 and 7,542 fixed disease panels, median Kendall tau-b was 1.0000 and 0.6248 across the two transitions, while leading-target sets changed in 6.21% and 17.38% of panels. Sensitivity persisted under strict entity stability, a broader disease universe and matched genetic-association analyses. Unique-leader replacement predominated over tie-induced variation. Independent recomputation and repeated clean builds reproduced all frozen outputs. The data record and software enable researchers to audit version-dependent rankings and distinguish score revision, reranking, roster turnover and semantic change when reusing dynamic biomedical knowledgebases.
