Recently, Sifei Han's team published a research paper titled “Predicting lymphatic transport potential using graph transformer based on limited historical data from in vivo studies” in the Journal of Controlled Release, a leading journal in the discipline of pharmacy. Predicting lymphatic transport potential using graph transformer based on limited historical data from in vivo studies”. Sifei Han from the School of Pharmacy, Tohoku University, and Di Wu from the University of Southern Queensland, Australia, are the co-corresponding authors. Visiting scholar Bambu Li, Master's student Ruiya Liu, class of 2024, School of Pharmacy, and Ph.D. student Zonghao Ji, class of 2023, National Supercomputing Center Jinan Branch, are the co-first authors. China Pharmaceutical University (CPU) was the first author and the first correspondent.
In recent years, drug targets located in the human lymphatic system are being gradually explored and utilized, bringing new opportunities for the treatment of tumors, infections, inflammation self-immunity, metabolic syndrome and other major diseases. Since blood is the dominant pathway for drug absorption and transportation, usually the amount of drug absorbed via lymph is extremely low (<0.2%), which is not conducive to its action within the lymphatic system. Previous studies have found that a small proportion of lipophilic drugs possess translymphatic absorption characteristics, but the accuracy of this empirical rule is limited, and there is an urgent need for more in-depth exploration of drug translymphatic absorption patterns and better predictive modeling in order to guide the molecular design and screening of drugs with high translymphatic absorption.
In the present work, we first searched the literature on drug lymphatic absorption measured by animal models (including human clinical trials), and constructed a database based on the differences in model selection. Aiming at the difficulty of accurate in vivo study of drug lymphatic absorption and the low overall data sample size (the number of drugs tested for lymphatic absorption is less than 5% of the total number of drugs), this work introduces a data augmentation model for AI data training under small sample conditions. Subsequently, based on the molecular structure code of compounds SMILES was selected as the best performing GT for lymphatic absorption prediction through the comparison of three modeling methods Graph Convolutional Network (GCN), Graph Attention Network (GAT) and Graph Transformer (GT) model. Finally, the model accuracy was validated with unpublished animal lymphatic transport data of in-lab drug testing.
This work is the world's first publication of an AI-driven drug uptake prediction model for the lymphatic system, providing a powerful tool for lymphotropic drug design and screening. The model is open for use online and is expected to be rapidly iterated with the increase of test data in the future. The group of Sifei Han and Luojuan Hu in the School of Pharmacy focuses on lymphatic drug transport and delivery systems, including work in the subfields of large molecule carrier construction, small molecule design, and lymphoid organoid microarrays. Since the establishment of the group in 2023, the team has collaborated extensively with more than ten groups inside and outside the university, and is promoting the utilization and clinical translation of drug targets within the lymphatic system.
The research work has been guided and assisted by Prof. Wei Qiang from the Institute for Original Drug Research of the University in database construction, Prof. Jiang Zhengyu and Prof. Jiang Cheng from the School of Pharmaceutical Sciences of the University in the design and synthesis of lymphoid-targeting molecular fragments, and Prof. Trevaskis from Monash University, Australia, in drug screening. This work was supported by the National Overseas Excellent Postdoctoral Introduction Program, and the Key Laboratory of Pharmaceutical Molecular Design and Optimization of Drug Formation in Jiangsu Province under Grant (DDORC202202).
Link to the paper: https://doi.org/10.1016/j.jconrel.2025.113847
Schematic diagram of the principle of AI-driven lymphatic prediction model based on data augmentation and graphical drug molecular structure
