Computer-aided design usually gives inspirations and has become a vital strategy to develop novel pesticides through reconstructing natural lead compounds. Patulin, an unsaturated heterocyclic lactone mycotoxin, is a new natural PSII inhibitor and shows significant herbicidal activity to various weeds. However, some evidence, especially the health concern, prevents it from developing as a bioherbicide. In this work, molecular docking and toxicity risk prediction are combined to construct interaction models between the ligand and acceptor, and design and screen novel derivatives. Based on the analysis of a constructed patulin–Arabidopsis D1 protein docking model, in total, 81 derivatives are designed and ranked according to quantitative estimates of drug-likeness (QED) values and free energies. Among the newly designed derivatives, forty-five derivatives with better affinities than patulin are screened to further evaluate their toxicology. Finally, it is indicated that four patulin derivatives, D3, D6, D34, and D67, with higher binding affinity but lower toxicity than patulin have a great potential to develop as new herbicides with improved potency.