Potential Molecular Links Between Benzophenone-3 and Breast Cancer: Network Toxicology, Machine Learning, Molecular Modeling, and Gene Expression Validation

Ziqi Wang,1 Yingluan Wang,2 Zhan Cao,3 Yi Jiang11Department of Breast Surgery, The Second Hospital of Jilin University, Changchun, 130041, People’s Republic of China; 2Department of Ophthalmology, The Second Affiliated Hospital of Harbin Medical University, Harbin, 150086, People’s Republic of China; 3Department of Neurology, The Second Affiliated Hospital of Harbin Medical University, Harbin, 150086, People’s Republic of ChinaCorrespondence: Yi Jiang, Department of Breast Surgery, The Second Hospital of Jilin University, Changchun, 130041, People’s Republic of China, Tel +86 15526891025, Email jyiedu@163.comPurpose: Benzophenone-3 (BP-3) is a widely used ultraviolet filter with potential carcinogenic concern, but its molecular link with breast cancer remains unclear. This study aimed to identify candidate targets and potential molecular links between BP-3 and breast cancer using toxicological prediction, machine learning, molecular modeling, and gene expression analysis.Methods: The ADMET and toxicity profiles of BP-3 were evaluated using PubChem, ADMETlab 3.0, ProTox 3.0, and pkCSM. BP-3-related targets were predicted from ChEMBL, STITCH, and SwissTargetPrediction, while breast cancer-related differentially expressed genes were obtained from GEO datasets. Candidate targets were screened by intersection analysis and examined using GO/KEGG enrichment and PPI networks. Core genes were identified using 127 machine learning model combinations and SHAP interpretation. GSVA, GSEA, TIMER2.0, molecular docking, molecular dynamics simulations, MM-GBSA calculations, and qRT-PCR were used for validation.Results: Seventy BP-3-associated candidate genes related to breast cancer were identified. Among 127 model combinations, Stepglm[both]+SVM showed the best performance, with an average AUC of 0.935, and identified MMP13, ERBB2, and HMGCR as core genes. Functional analyses suggested that these genes were mainly involved in extracellular matrix remodeling, lipid metabolism reprogramming, and energy metabolism regulation. Molecular docking predicted favorable binding poses between BP-3 and the candidate proteins, with docking scores ranging from − 5.8 to − 8.5 kcal/mol. Molecular dynamics and MM-GBSA analyses indicated stable simulated complexes, with calculated binding free energies below − 14.30 kcal/mol. qRT-PCR showed higher expression of MMP13, ERBB2, and HMGCR in the examined breast cancer cell lines than in normal mammary epithelial cells.Conclusion: Integrated computational analyses identified MMP13, ERBB2, and HMGCR as candidate genes potentially linking BP-3 with breast cancer-related molecular features. Molecular docking and molecular dynamics simulations predicted feasible interactions between BP-3 and the corresponding proteins, while qRT-PCR confirmed the differential expression of these genes in breast cancer cell lines. These findings represent computational predictions and correlational evidence rather than proof of direct BP-3 regulation or causation. Direct BP-3 exposure experiments in cellular and animal models are required to establish causal relationships.Keywords: network toxicology, molecular docking, machine learning, SHAP, molecular dynamics, breast cancer