Integration of network toxicology, machine learning and single-cell sequencing identifies candidate molecular links between air pollutants and hepatocellular carcinoma.

Abstract

Background: Epidemiological studies link long-term air pollution to an increased risk of hepatocellular carcinoma (HCC), but the underlying toxicological targets remain poorly understood. We used an integrative computational framework to identify and prioritize candidate molecular mediators potentially linking pollutant-associated gene signatures with hepatocarcinogenesis.

Methods: We retrieved pollutant-responsive genes associated with seven toxicants from the Comparative Toxicogenomics Database and intersected them with HCC-associated genes. We used protein-protein interaction (PPI) network analysis to identify hub nodes. An optimized machine learning pipeline (glmBoost and Ridge regression) was trained on GSE36376 and validated in three independent cohorts. We mapped gene program activity within the tumor microenvironment using single-cell RNA sequencing (scRNA-seq) and modeled perturbations with scTenifoldKnk. Finally, we assessed structural compatibility between pollutants and protein targets.

Results: We identified 240 genes at the intersection of pollutant and HCC sets. Enrichment analysis highlighted innate immune signaling and chronic inflammation, specifically the IL-17, TNF, and NF-κB pathways. Hub nodes included IL6, TNF, MMP9, and AKT1. The machine learning model prioritized five candidate transcriptomic markers: FOS (AUC = 0.944), PARP1 (AUC = 0.929), MMP9 (AUC = 0.813), CCL5 (AUC = 0.747), and JUN (AUC = 0.733), all validated across cohorts. scRNA-seq module scoring showed the highest activity in T cells, monocytes, and macrophages. However, these data lack individual-level pollutant exposure documentation. In silico MMP9 knockout perturbed genes involved in angiogenesis and the extracellular matrix. Molecular docking suggested preliminary structural compatibility between toluene and MMP9 (-5.2 kcal/mol).

Conclusion: This analysis outlines a potential "pollutant-immune/inflammation-HCC" framework and identifies five transcriptomic signatures for further validation. These findings support a hypothesis that air pollutant-associated molecular programs may be linked to HCC pathobiology and warrant experimental validation.