Mixture-of-Experts\nBased Dissociation Kinetic Model\nfor <i>De Novo</i> Design of HSP90 Inhibitors with Prolonged\nResidence Time

The dissociation rate constant (<i>k</i><sub>off</sub>) significantly impacts the drug potency and dosing frequency.\nThis\nwork proposes a powerful optimization-based framework for <i>de novo</i> drug design guided by <i>k</i><sub>off</sub>. First, a comprehensive database containing 2,773 unique <i>k</i><sub>off</sub> values is created. Based on the database,\na novel generic dissociation kinetic model is developed with a mixture-of-experts\narchitecture, enabling high-throughput predictions of <i>k</i><sub>off</sub> with high accuracy. The developed model is then integrated\nwith an optimization-based mathematical programming approach to design\ndrug candidates with low <i>k</i><sub>off</sub>. Finally,\nthe τ-RAMD method is utilized to rigorously verify the designed\npotential drug candidates. In a case study, the framework successfully\nidentified numerous new potential HSP90 inhibitor candidates, achieving\na maximum 45.7% improvement in residence time (τ = 1/<i>k</i><sub>off</sub>) compared to that of a known exceptional\nHSP90 inhibitor. These findings demonstrate the feasibility and effectiveness\nof the kinetics-guided optimization-based <i>de novo</i> drug design framework in designing drug candidates with prolonged\nτ.