.png){kind=icon}
.png){kind=icon}
Return
FEMTL-DR: A feature-enhanced multi-task learning model for flexible drug recommendation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.131806.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Multiple disease-based syndrome reasonings and their corresponding drug recommendations are crucial for personalized diagnosis and treatment in Traditional Chinese Medicine (TCM). However, it remains a challenging task to effectively extract and integrate various entities and multi-dimensional relationships for syndrome-based drug recommendations. This study investigates FEMTL-DR, a novel feature-enhanced multi-task learning model for flexible drug recommendation. Based on various entity characteristics, we propose a hybrid multi-entity encoding strategy to realize diverse disease, syndrome, herb, and herb property encodings. Then, the pairwise similarity between different entity encodings is calculated to reconstruct the adjacency matrix of the heterogeneous graph. A state-space transformer-based strategy is presented to enhance node features by capturing long-range dependencies and extracting local and global information. Finally, a multi-task learning framework combined with a graph neural network and transformer module is constructed to learn the relationships between enhanced node features for syndrome classification and drug recommendation. Taking Reflux Esophagitis (RE) as an instance, a series of experiments have been conducted to verify the performance of the proposed FEMTL-DR, including comparison experiments, ablation verification, and parameter sensitivity tests. The experimental results indicate that the proposed model outperforms baselines on representative evaluation metrics. Specifically, the overall evaluation (integrating syndrome classification and drug recommendation) achieves improvements of at least 2.48% in Average Precision (AP), 2.93% in Precision, 2.77% in Recall, and 3.08% in F1-score. This study provides a novel solution for personalized diagnosis and treatment for RE in TCM and can be considered a paradigm for other diseases.
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W
View more organizations