Enhancing Precision Drug Recommendations via Fine-grained Exploration of Motif Relationships
Keywords: Drug discovery, Recommender System, Molecular Representation Learning
Abstract: Making accurate and safe drug recommendation for patients has always been a challenging task. Even though rule-based, instance-based, and longitudinal data-based approaches have made notable strides in drug modeling, they often neglect to fully leverage the rich motifs information. However, it is widely acknowledged that motifs exert a significant influence on both drug action and patient symptomatology. Therefore, there is a pressing need for more comprehensive exploration this invaluable information to further enhance drug recommendation systems. To tackle the aforementioned challenges, we present DEPOT, a novel drug recommendation framework that leverages motifs as higher-level structures to enhance recommendations. In our approach, we employ chemical decomposition to partition drug molecules into motif-trees, enabling us to capture the structural information among substructures. To investigate the relationship between disease progression and motifs, we conduct a meticulous exploration from two perspectives: repetition and exploration. This comprehensive analysis allows us to gain valuable insights into the drug turnover, with the former focusing on reusability and the latter on discovering new requirements. Furthermore, we incorporate historical DDI effects and employ a nonlinear optimization objective to stabilize the training process, ensuring the safety of recommended drug combinations. Extensive experiments are conducted on two data sets to validate the uniqueness of the DEPOT framework and the efficacy of the individual submodules.
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3516
Loading