AI-guided design of self-assembled flavonoid-cisplatin nanoparticles enhances triple-negative breast cancer therapy via three cell-death pathways

Abstract

Triple-negative breast cancer (TNBC) remains challenging to treat because of its aggressiveness and poor response to chemotherapy. Although cisplatin is clinically used for TNBC, its therapeutic window is narrow due to severe systemic toxicity and rapid development of drug resistance, underscoring the need for new strategies. To address this, we developed LMCDPS, a molecular language-model platform that learns the structural grammar of natural compounds, predicts their compatibility for co-assembly with cisplatin, and uniquely traces active small molecules back to their botanical origins. Using LMCDPS, we identified persimmon leaf-derived flavonoids as optimal cisplatin partners and confirmed their ability to spontaneously form stable, excipient-free nanoparticles (PLF-Cis NPs) via π–π interaction and hydrogen bonding. These nanoparticles eliminate synthetic carriers, reduce impurities in crude extracts, and substantially enhance tumor delivery, achieving an 11.3-fold increase in intracellular platinum accumulation. PLF-Cis NPs exert potent anti-TNBC activity by coordinating cisplatin-induced apoptosis, flavonoid-mediated ferroptosis, and immunogenic cell death, thereby promoting dendritic-cell maturation and robust CD8⁺ T-cell infiltration. In an orthotopic 4T1 model, they achieved a 71.5% reduction in tumor growth. This study establishes a language-model-driven framework for designing natural product-based, excipient-free nanomedicines, offering a scalable path to enhance chemotherapy while mitigating systemic toxicity.