Distribution Matching for Drug-Drug Interaction Prediction

EasyChair Preprint 14299

Abstract

Drug-Drug interactions (DDIs) refer to the mutual effects that may occur when two or more drugs are co-administered. These interactions have significant implications for the efficacy, safety, or tolerability of medications, making them of paramount importance in the field of medicine. Recent research has focused on deep learning techniques, such as graph-based learning methods, which typically consider the molecular structure information of drugs but often neglect inter-view information.To overcome this limitation, we propose a hierarchical graph-based deep learning method that efficiently aligns intra-view and inter-view embeddings. Specifically, we perform distribution matching in both the feature space and output space, maximizing mutual information between the two views to enhance prediction accuracy. Additionally, we introduce a novel loss function that uses central matching distribution (CMD) to balance information between intra-view and inter-view embeddings, instead of relying on unsupervised contrastive learning. This approach increases computational speed by 50\% while maintaining high accuracy. Evaluations on three datasets demonstrate that our method outperforms other state-of-the-art models in DDI prediction.

Keyphrases: Hierarchical graph representation, central matching distribution, drug-drug interaction, graph embedding

@booklet{EasyChair:14299,
  author    = {Olivia Lee and Su Yu},
  title     = {Distribution Matching for Drug-Drug Interaction Prediction},
  howpublished = {EasyChair Preprint 14299},
  year      = {EasyChair, 2024}}