Reproduce “Natural Language-Assisted Multi-modal Medication Recommendation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jie Tan, Yu Rong, Kangfei Zhao, Tian Bian, Tingyang Xu, Junzhou Huang, Hong Cheng, Helen Meng

NLA-MMR is a framework for combinatorial medication recommendation that integrates textual Electronic Health Record (EHR) data and medication descriptions with chemical structural information using Pre-trained Language Models and Graph Neural Networks. The system achieves improved recommendation accuracy and better alignment with actual drug-drug interaction rates on MIMIC-III, MIMIC-IV, and eICU datasets.

Key insights

  • Treating patient and medication information as distinct modalities and explicitly aligning them in a shared semantic space can enhance recommendation accuracy and safety.
  • Leveraging natural language descriptions from both patient EHRs and medication databases via Pre-trained Language Models (PLMs) extracts rich, previously underutilized clinical knowledge.
  • Combining chemical structural information (via GNNs) with functional textual descriptions (via PLMs) provides more comprehensive and clinically relevant medication embeddings.
  • Integrating historical medication data through an attention mechanism improves the model's ability to capture patient trajectories over time.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.