Identification of four‑gene signatures overlapping between predicted TDCPP targets and breast cancer transcriptomic features via network toxicology and machine learning

Tris(1,3-dichloro-2-propyl) phosphate (TDCPP) is a chlorinated organophosphate flame retardant widely detected in indoor products and environmental media. Existing evidence suggests that TDCPP may have endocrine-disrupting, carcinogenic, and immune-related effects. However, whether predicted TDCPP-related molecular targets overlap with breast cancer-associated transcriptomic features remains unclear. This study used an integrated network toxicology and transcriptomic approach to identify candidate shared molecular features between predicted TDCPP-related targets and breast cancer. TDCPP-related targets were collected from ChEMBL, CTD, and PharmMapper. These targets were intersected with breast cancer-related genes and differentially expressed genes from the TCGA-BRCA cohort. Weighted gene co-expression network analysis, protein–protein interaction analysis, GO and KEGG enrichment analyses, and three machine learning algorithms were used to screen candidate hub genes. Diagnostic modeling, SHAP analysis, immune infiltration analysis, and molecular docking were then used to evaluate the predictive and biological relevance of these genes within the computational framework. The analysis identified 4,888 potential TDCPP-related targets and 2,255 breast cancer-related genes, with 706 overlapping targets. By integrating differentially expressed genes, TDCPP–BRCA common targets, and the key WGCNA module, 89 candidate genes were selected. These genes were mainly enriched in chemical response, cell proliferation, signal transduction, and cancer-related pathways. Machine learning analysis identified four hub genes: ANXA1, DMD, FGF1, and ADH1C. The four-gene model showed strong separation between tumor and normal samples in the TCGA-BRCA dataset, with an AUC of 0.999. However, this performance should be interpreted within the analyzed cohort and requires external validation. SHAP and permutation importance analyses suggested that these genes contributed to model prediction, especially ADH1C and ANXA1. Immune infiltration analysis showed that the hub genes were associated with several immune cell types, including T cells, macrophages, and dendritic cells. Molecular docking suggested possible interactions between TDCPP and the four hub proteins, with predicted binding energies ranging from − 5.0 to − 6.3 kcal·mol⁻¹. These findings identify ANXA1, DMD, FGF1, and ADH1C as candidate genes within a computationally inferred overlap network between predicted TDCPP-related targets and breast cancer-associated transcriptomic changes. Functional enrichment, immune infiltration, and docking analyses suggest that these genes may be related to pathways involving inflammation, growth-factor signaling, metabolism, cytoskeletal regulation, and immune microenvironment variation. Because TCGA-BRCA does not contain measured TDCPP or BDCPP exposure data, these results should be interpreted as hypothesis-generating rather than evidence of a direct exposure–disease relationship. Further validation using exposure-based epidemiological data, experimental breast cancer models, and targeted TDCPP–protein interaction assays is needed.