Drug-drug interaction analysis based on information bottleneck graph neural network: A review.

The objective of learning drug-drug interactions is to understand the interaction behavior between compound molecules, which has garnered significant interest in the field of compound molecular science due to the potential harm adverse drug interactions may cause to organisms. Existing machine learning methods mostly rely on manually designed representations of compound molecules, overlooking the essence of compound molecules being composed of multiple molecular substructures and constrained by the knowledge of domain experts in the field of compound molecules. In this work, we propose a novel graph neural network framework for learning compound molecule interactions, which investigates the relationship between pairs of compound molecule graphs by detecting core molecular subgraphs of compound molecules. The proposed graph neural network learning framework leverages the fundamental principle of conditional graph information bottleneck to find the minimum information containing molecular subgraph for a given pair of compound molecule graphs. This framework effectively predicts the essence of compound molecule reactions, wherein the core structure of a compound molecule depends on its interaction with other compound molecules. Extensive experiments on common datasets for prediction tasks of compound molecule interactions demonstrate that the proposed graph neural network learning framework enhances the predictive performance of compound molecule interactions.