An integrated environmental toxicity risk assessment framework combining deep learning and molecular simulation: A case study on pyrethrins and breast cancer.

This study develops and validates a multi-scale integrative computational toxicology framework to systematically investigate the potential association between the natural pesticides pyrethrin I and II and breast cancer risk. The proposed approach integrates deep learning-based drug-target interaction prediction (via DeepPurpose), molecular docking and dynamics (MD) simulations, and protein-protein interaction (PPI) network modeling, forming a traceable risk inference chain from molecular-level interactions to clinically relevant outcomes. Experimental results revealed that pyrethrin I exhibits stable and high-affinity binding to key breast cancer-related proteins, including RPS6KB1, TNKS2, and MAOB, with a minimum binding free energy (ΔG) reaching -27.37 kcal/mol. These interactions potentially modulate tumor progression through key oncogenic pathways such as PI3K/AKT, Wnt/β-catenin, and metabolic reprogramming. RPS6KB1 is implicated in estrogen receptor-positive (ER+) breast cancer proliferation, while TNKS2 is closely associated with stemness maintenance and the aggressiveness of triple-negative breast cancer (TNBC). MAOB demonstrates the highest structural stability among complexes, making it a promising candidate for toxicological modeling. The study further introduces a cross-scale risk indicator modeling strategy, constructing a mechanistically interpretable chain from compound structure to protein modules, carcinogenic pathways, and clinical risks. This integrative methodology supports environmental exposure monitoring and toxicological policy development. The open-source, containerized analytical toolchain developed herein is highly extensible and adaptable for future toxicological evaluations of other natural compounds and emerging environmental pollutants.