ReproTox-KG

ReproTox-KG: Knowledge Graph for Drug–Gene–Birth Defect Associations

ReproTox-KG integrates heterogeneous biomedical datasets to model and predict associations between small molecule compounds, genes, and structural birth defects using a knowledge graph framework.

Key Features:

• Data Integration: Consolidates drug/birth-defect co-mentions from published abstracts, gene/birth-defect associations from genetic studies, drug- and preclinical compound-induced gene expression changes in cell lines, known drug targets, genetic burden scores for human genes, and placental crossing scores for small molecules.
• Semi-Supervised Learning (SSL): Applies SSL to evaluate over 30,000 preclinical small molecules for predicted placental transfer and teratogenic potential using labeled and unlabeled data within the knowledge graph.
• Clique Identification: Detects over 500 birth-defect/gene/drug cliques to characterize putative molecular mechanisms underlying drug-induced structural birth defects.

Scientific Applications:

• Teratogenicity Assessment: Identifies potential teratogenic risks of small molecule compounds, investigates genetic contributors to structural birth defects, and supports prediction of adverse developmental effects during pregnancy.

Methodology:

Constructs a knowledge graph integrating literature-derived associations, genetic study data, gene expression perturbation profiles, drug target annotations, genetic burden metrics, and placental crossing scores. Applies semi-supervised learning to propagate labels across graph nodes and identify high-confidence drug–gene–birth defect associations, followed by clique detection to reveal mechanistic subgraphs.