Rational design of low-risk and high-efficiency β-triketone herbicide alternatives via integrative computational toxicology and NGBoost modeling.

β-Triketone herbicides (bTHs) are widely applied in crops such as maize, rice, and sorghum. However, their efficacy on some gramineous weeds is limited, and concerns over their persistence and toxicity are rising. In this study, a comprehensive evaluation framework integrating Box-Cox transformation and coefficient of variation methods was developed to simultaneously characterize herbicidal efficacy, soil toxicity, and aquatic toxicity of bTHs. Based on this framework, a 3D-QSAR model describing the combined effects of biological toxicity and herbicidal activity was constructed. Guided by the 3D contour maps, 32 bTH derivatives were rationally designed using tembotrione (TEM) as the lead compound, exhibiting improved comprehensive performance (32.98%-74.73%). Specifically, soil and aquatic toxicity were reduced by 36.42%-38.56% and 38.32%-47.87%, respectively. Further prediction using EPI Suite indicated reduced aquatic toxicity. By constructing a machine learning model based on NGBoost, we observed enhanced biodegradability (14.23%-54.33%). Two candidate compounds (T-2 and T-32) were identified as promising molecules with predicted higher herbicidal efficacy and lower overall toxicity. Gaussian-based mechanistic analysis suggested that the β-diketone moiety of these derivatives enhanced electron delocalization and stronger σ-donor capability. This may stabilize the Fe²⁺ coordination complex in the HPPD active site and thereby improve inhibitory activity. Meanwhile, substituent modifications regulated charge distribution and molecular conformation. This may reduce nonspecific interactions with toxicity-related functional groups in non-target organisms, thus lowering biological toxicity. This study provides a rational design strategy to simultaneously improve herbicidal efficacy and reduce the environmental and health risks of bTHs, theoretically guiding green pesticide development. These computational findings identify promising bTH candidates and a rational design approach, which warrant further experimental validation.