Tong, J.; Wang, X.; Lu, M.; Wang, T.; Chen, J.; Chen, J.; Li, Y.; Xie, C.; Fu, Y.; Yu, C.

Radioresistance remains a primary obstacle in tumor radiotherapy, yet no radiosensitizers have been approved for clinical use due to toxicity concerns. To identify effective and safe radiosensitizers, a natural products (NP) database comprising 79,263 compounds were docked against a hybrid target library of four DNA damage repair (DDR)-related proteins, including both experimental and AI-predicted structures generated by AlphaFold3 and RoseTTAFold All-Atom models. It demonstrated that hit compounds identified by AI-modeled structures accounted for 80% of all hits in ataxia telangiectasia mutated (ATM), 88.89% in ATM- and Rad3-related (ATR), 92.31% in DNA-dependent protein kinase catalytic subunit (DNA-PKcs), and 70% in Poly (ADP-ribose) polymerase 1 (PARP1), expanding compounds diversity recognized by experimental structures. Seven promising multi-target radiosensitizer candidates were identified, among which two compounds were validated as effective radiosensitizers in vitro. Detailed proteomic analyses revealed that both compounds activate TP53-associated apoptosis and senescence as cellular endpoints by modulating the synergistic interplay between DDR and spindle checkpoints. The integration of AI-predicted and experimental structures highlighted the structural complementarity advantages of this approach for small-scale compound libraries, where limited computational resources are required. This study provided new radiosensitizers and strategies to overcome tumor radioresistance in future research.