MMM: Quantum-Chemical Molecular Representation Learning for Combinatorial Drug Recommendation

TL;DR

MMM addresses the challenge of drug–drug interactions by fusing 3D quantum-chemical ELF maps with longitudinal EHR data in a multimodal architecture. The framework uses an ELF-based drug encoder to capture global electronic properties and a local bipartite encoder to model substructure-level interactions, enabling safer, personalized drug recommendations. On MIMIC-III, MMM outperforms graph-based baselines in DDI reduction and therapeutic prediction, with statistically significant gains, and ablation confirms the complementary value of its components. This approach offers a chemically informed, clinically actionable pathway for safer combinatorial prescribing, while suggesting future work to incorporate richer mechanistic DDI information and broader datasets.

Abstract

Drug recommendation is an essential task in machine learning-based clinical decision support systems. However, the risk of drug-drug interactions (DDI) between co-prescribed medications remains a significant challenge. Previous studies have used graph neural networks (GNNs) to represent drug structures. Regardless, their simplified discrete forms cannot fully capture the molecular binding affinity and reactivity. Therefore, we propose Multimodal DDI Prediction with Molecular Electron Localization Function (ELF) Maps (MMM), a novel framework that integrates three-dimensional (3D) quantum-chemical information into drug representation learning. It generates 3D electron density maps using the ELF. To capture both therapeutic relevance and interaction risks, MMM combines ELF-derived features that encode global electronic properties with a bipartite graph encoder that models local substructure interactions. This design enables learning complementary characteristics of drug molecules. We evaluate MMM in the MIMIC-III dataset (250 drugs, 442 substructures), comparing it with several baseline models. In particular, a comparison with the GNN-based SafeDrug model demonstrates statistically significant improvements in the F1-score (p = 0.0387), Jaccard (p = 0.0112), and the DDI rate (p = 0.0386). These results demonstrate the potential of ELF-based 3D representations to enhance prediction accuracy and support safer combinatorial drug prescribing in clinical practice.

Figures (1)

• Figure 1: Proposed model architecture. We first encode longitudinal EHRs into patient state vectors, which are then used to compute global and local drug representations via an ELF-based encoder and a bipartite substructure encoder. The two drug vectors are fused to generate safe and personalized drug recommendations.