Objective To predict and analyze the potential molecular mechanisms by which polyethylene terephthalate (PET) may induce depression, based on computational toxicology, machine learning, and molecular docking strategies.

Methods The peripheral blood transcriptome datasets of depression (GSE52790, GSE38206, and GSE76826) were downloaded and integrated from the GEO database. Through differential expression analysis (|log₂FC|> 0.25, P_adj < 0.05) and weighted gene co-expression network analysis (WGCNA), 645 depression-related genes were identified. Potential PET targets (n=567) were predicted using the SwissTargetPrediction, SEA, and PharmMapper databases. Intersection with depression-related genes yielded 32 common targets. A protein-protein interaction (PPI) network was constructed, and GO and KEGG enrichment analyses were performed. Seven machine learning algorithms were used to construct 110 models; the elastic net (Enet) model showed the best performance (highest area under the curve AUC), and key genes were interpreted using SHAP. Molecular docking was performed between PET and the core genes (ADORA2B, JAK2, PYGL, and RAN) for validation.

Results The 32 intersecting targets were mainly enriched in pathways related to Th17/Th1/Th2 cell differentiation, necroptosis, intrinsic apoptotic signaling, and regulation of humoral levels. Machine learning analysis identified PYGL, JAK2, RAN, and ADORA2B as the most predictive feature genes; among them, PYGL achieved an AUC of 0.817 in an independent validation dataset. Molecular docking showed that the binding energies between PET and the four targets ranged from -8.9 to -7.8 kcal/mol, indicating stable binding.

Conclusion From the perspective of computational toxicology, this study suggests that PET microplastics may induce depression by interacting with key molecules, such as JAK2, ADORA2B, PYGL, and RAN.