Drug-drug interactions (DDIs) represent a critical challenge in pharmacology, often leading to adverse drug reactions with significant implications for patient safety and healthcare outcomes. While graph-based methods have achieved strong predictive performance, most approaches treat drug pairs independently, overlooking the complex, context-dependent interactions unique to drug pairs. Additionally, these models struggle to integrate biological interaction networks and molecular-level structures to provide meaningful mechanistic insights. In this study, we propose MolecBioNet, a novel graph-based framework that integrates molecular and biomedical knowledge for robust and interpretable DDI prediction. By modeling drug pairs as unified entities, MolecBioNet captures both macro-level biological interactions and micro-level molecular influences, offering a comprehensive perspective on DDIs. The framework extracts local subgraphs from biomedical knowledge graphs and constructs hierarchical interaction graphs from molecular representations, leveraging classical graph neural network methods to learn multi-scale representations of drug pairs. To enhance accuracy and interpretability, MolecBioNet introduces two domain-specific pooling strategies: context-aware subgraph pooling (CASPool), which emphasizes biologically relevant entities, and attention-guided influence pooling (AGIPool), which prioritizes influential molecular substructures. The framework further employs mutual information minimization regularization to enhance information diversity during embedding fusion. Experimental results demonstrate that MolecBioNet outperforms state-of-the-art methods in DDI prediction, while ablation studies and embedding visualizations further validate the advantages of unified drug pair modeling and multi-scale knowledge integration. Moreover, the pooling strategies yield interpretable insights into the underlying interaction mechanisms, highlighting MolecBioNet's potential as a robust and transparent framework for predictive modeling in support of drug safety assessment.

In recent years, combined drug screening has played a very important role in modern drug discovery. Generally, synergistic drug combinations are crucial in treatment for many diseases. However, the toxic side effects of drug combinations are probably increased with the increase of drugs numbers, so the accurate prediction of toxic side effects of drug combinations is equally important. In this paper, we built a Metapath-based Aggregated Embedding Model on Single Drug-Side Effect Heterogeneous Information Network (MAEM-SSHIN), which extracts feature from a heterogeneous information network of single drug side effects, and a Graph Convolutional Network on Combinatorial drugs and Side effect Heterogeneous Information Network (GCN-CSHIN), which transforms the complex task of predicting multiple side effects between drug pairs into the more manageable prediction of relationships between combinatorial drugs and individual side effects. MAEM-SSHIN and GCN-CSHIN provided a united novel framework for predicting potential side effects in combinatorial drug therapies. This integration enhances prediction accuracy, efficiency, and scalability. Our experimental results demonstrate that this combined framework outperforms existing methodologies in predicting side effects, and marks a significant advancement in pharmaceutical research.

Drug combinations are vital for treating complex diseases and advancing drug development, but accurately identifying synergistic combinations remains a significant challenge. Although graph neural networks (GNNs) have recently been used to predict drug combinations, the complex interactions between drugs and multimodal data (e.g., target proteins) and the prevalent high-order relations among drugs have yet to be fully exploited. The hypergraph offers a natural methodology for modeling high-order relations and provides profound insights for multimodal fusion. Here, we introduce the multimodal relational hypergraph neural network (MRHGNN), a novel framework for predicting synergistic drug combinations. Specifically, we design a dual-channel architecture to capture the physicochemical attributes of drugs and their interactive synergies, thereby facilitating the generation of multimodal drug representations. To obtain comprehensive representations of drugs, we use an attention mechanism to explore complementarity among multimodal drug embeddings. In addition, the unified framework jointly learns primary and self-supervised learning tasks, fostering a robust predictive capability. Experimental results demonstrate that MRHGNN accurately predicts synergistic drug combinations, and the effectiveness of the dual-channel setup and motif structures has been validated through ablation studies. Further literature searches illustrate that our model holds significant promise in accelerating the discovery of novel synergistic drug combinations, particularly in cancer therapy. This study not only introduces a novel computational tool but also paves the way for advanced methodologies in drug discovery and development.

When facing choices such as getting vaccinated, traveling, and going to the cinema, individuals can be classified into two groups based on their decision-making strategy: 1) selfish agents, who make decisions to maximize their own payoff; and 2) collectivist agents, who try to maximize the group’s benefits. We model both strategies in the evolutionary dynamics of signed networks to examine how both types of agents choose to behave in society. In particular, we run simulations and find that when agents’ strategies are fixed, collectivist agents or group decision-making are conducive to cooperation. Some network properties such as having structural balance and a clear modular organization promote cooperation. However, when strategic learning is considered, the dynamics become highly uncertain. When collectivist agents are initially located in a closely related group, the cooperation ratio and proportion of collectivist agents both rise. These observations reveal that besides network properties, the topology distribution of agents with different decision-making tendencies plays a significant role in social evolution.