Lab-in-the-loop machine learning for brain-targeting delivery system design
Qiujun Qiu, Shiyi Li, Jixian Zhang, Jixiang Chen, Xinyi Ding, Shengyao Liu, Jianyong Sheng, Zhiqing Pang, Ru Zhang, Anni Wang, Meichen Dong, Meng Zhang, Miaomiao Zhang, Tun Lu, Ning Gu, Shuigeng Zhou, Defang Ouyang, Dongsheng Li, Shuangjia Zheng, Jianxin Wang
Abstract
Nanomedicine-based drug delivery systems (DDSs) show promise in boosting brain drug accumulation, but the key factors affecting their distribution remain unclear. AI algorithms capable of processing complex biological data could decipher the relationship between brain physiology and DDSs, though such advanced methods are not yet available. Herein, we applied the lab-in-the-loop framework to analyze and design brain-targeting DDSs, combining 17,600 features from 9,500 publications to train specialized machine-learning models. The best-performing algorithm outperforms conventional statistical predictions and identified particle size and zeta potential as critical determinants of brain delivery efficiency. We further predicted and visualized delivery efficiencies for a large number of unexplored DDSs in a multidimensional feature space and successfully designed several high-efficiency DDSs through Bayesian optimization. Our study offers a large-scale open access resource to advance DDS distribution understanding and may guide the development of brain-targeting DDSs.