A machine-learning-guided hydrogen-bonded organic framework for long-term, ultrasound-triggered pain therapy

Effective treatment of chronic pain remains hindered by the lack of drug delivery systems that simultaneously achieve long-term stability, high spatial precision, and non-invasiveness ^1,2,3 . Here, we utilize a programmable, ultrasound-responsive drug delivery platform based on hydrogen-bonded organic framework (HOF) nanoparticles, enabling on-demand anesthetic release with long-term and durable analgesic efficacy. A machine learning (ML)-guided screening pipeline was developed to evaluate about 250 FDA-approved drugs, spanning both hydrophilic and lipophilic agents, and identified bupivacaine (lipophilic) and lidocaine hydrochloride (hydrophilic) as optimal candidates. These agents were efficiently encapsulated into HOF nanoparticles via diffusion and double-solvent methods. Ultrasound-triggered drug release in vitro transiently suppressed calcium signaling in adeno-associated virus (AAV)-transfected neurons. In vivo , ultrasound-activated release of bupivacaine or lidocaine HCl at the sciatic nerve site significantly elevated mechanical nociceptive thresholds for up to seven days, reduced autotomy (self-mutilaion) behavior, and improved motor function in a rat model of chronic pain. Notably, the machine learning-identified top candidates not only exhibited high loading efficiency, but also demonstrated superior therapeutic outcomes in vivo , establishing a direct link between computational prediction and biological efficacy. This non-invasive, programmable HOF-based system provides a clinically translatable platform for on-demand, spatiotemporally precise pain management.