Deep learning-guided ligand generation for the strigolactone receptor ShHTL7

Striga hermonthica is a root-parasitic weed that poses a significant threat to crop production in sub-Saharan Africa. The strigolactone receptor ShHTL7, which mediates host-induced seed germination, is a potential target for the development of selective chemical regulators. Here, we present an integrated deep learning-assisted computational workflow for the discovery of ShHTL7-targeted ligands. Using the REINVENT4 platform combined with transfer learning, a ligand-generation model was constructed and coupled with a multi-parameter screening strategy, including physicochemical properties, ADMET-related descriptors, molecular docking, and molecular dynamics simulations. From the generated compounds, six candidates were prioritized for further evaluation. Docking analysis indicated that several candidates displayed favorable predicted interactions with the ShHTL7 binding pocket. Molecular dynamics simulations suggested stable conformational behavior of the selected ligand-protein complexes over the simulated timescale. Notably, inh-117 exhibited favorable binding energetics and broader residue-level contributions in MM/PBSA analysis compared with the parent ligand KK023. This study provides a computational framework for the prioritization of ShHTL7-targeted ligands and may guide future experimental efforts toward selective Striga regulators.