From Data to Mechanism: A Knowledge Graph-Bayesian Network-Driven Suspect Screening (KGBS) Strategy for Phthalate Exposure Assessment

The widespread substitution of phthalate esters (PAEs) in industrial production has resulted in an increasingly complex environmental exposure and metabolic transformation in humans. To systematically elucidate their biotransformation relationships, a knowledge graph-Bayesian network-driven suspect screening (KGBS) strategy that integrated text mining, probabilistic reasoning, and high-resolution mass spectrometry (HRMS) was developed. A total of 54,838 associations between PAEs and their metabolites were extracted through text mining from 3167 publications to construct a Bayesian inference-embedded knowledge graph, which quantitatively prioritized biotransformation events. This framework was then coupled with an HRMS workflow and a mechanistically informed spectral library, forming a unified KGBS platform for compound discovery. Applied to paired indoor dust and human urine samples, the KGBS framework identified 68 PAEs and 49 metabolites, including 18 PAEs and 14 metabolites, that were newly annotated in this study. The prioritization scores and diagnostic MS 2 fragmentation fingerprints provided mechanistically informed support for their metabolic linkages. By coupling Bayesian inference with suspect screening, KGBS enabled pathway-centric interpretation of contaminants and their transformation products, which established a scalable and self-evolving strategy for data-driven reconstruction of biotransformation networks, thus offering new perspectives for exposome characterization and human exposure risk assessment.