Network Toxicology and Machine Learning Prioritise a Seven-Gene Signature of Liver Injury Associated with Particle-Bound Polycyclic Aromatic Hydrocarbons

As a major toxic component of environmental fine particulate matter (PM_2.5), polycyclic aromatic hydrocarbons (PAHs) pose a significant risk to human health, particularly for liver damage, following inhalation or ingestion. However, their comprehensive molecular mechanisms underlying this hepatotoxicity have long remained unclear. In this study, we developed an integrative framework that combines network toxicology, batch-corrected two-stage machine learning, gene set variation analysis, immune-signature analysis and molecular docking. We retrieved 479 overlapping candidates from 570 PAH-related targets and 12,072 liver-injury-associated genes; the intersection of network hubs (MCC top-30 and MCODE modules) with differentially expressed genes across three merged hepatic cohorts (n = 174) yielded 32 candidates, from which three complementary algorithms (LASSO, SVM-RFE and permutation-calibrated random forest) distilled a seven-gene signature (CYP2C19, MYC, EGR1, HMMR, ICAM1, MAPK3 and NFE2L2). An eight-classifier comparison selected a support vector machine, exact Shapley values quantified each gene’s contribution, and a logistic risk score built on the seven genes discriminated metabolic dysfunction-associated steatotic liver disease (MASLD) with an area under the curve (AUC) of 0.953 in the training set, 0.828 in a held-out test set and 0.978 in an independent 216-biopsy validation cohort (10 controls and 206 MASLD cases), in which the score was discriminative but its probabilities were not well calibrated. GSVA and immune-signature analyses positioned the signature within a three-layer working model: collapse of metabolic and antioxidant capacity (CYP2C19, NFE2L2), imbalance of stress signalling (elevated MAPK3 with attenuated MYC and EGR1), and cell-behaviour remodelling (HMMR-linked proliferation programmes and ICAM1-linked immune remodelling), consistent with a state of heightened susceptibility to chemical stressors such as PAHs. In summary, this study provides a prioritised, hypothesis-generating target set and a reusable, batch-corrected analytical workflow for environmentally associated liver injury, with experimental validation of the prioritised targets as the necessary next step.