IDEAL (Interpretability-Driven Evolvable Attentive Learning for Microbe-Drug Association) is proposed, a multi-view framework that integrates drug network topological attributes, BERT-encoded drug semantics, drug fingerprints, microbe genome sequence attributes, BERT-encoded microbe semantics, and microbe metabolic pathway attributes.
Abstract
Predicting associations between human microbes and drugs (MDA) is a critical step in drug development and precision medicine. Although various computational approaches have been proposed, many existing models still struggle to reveal the key features and interaction mechanisms that drive their predictions, and their decision processes remain difficult to interpret and validate biologically. To address these limitations, we propose IDEAL (Interpretability-Driven Evolvable Attentive Learning for Microbe-Drug Association), a multi-view framework that integrates drug network topological attributes, BERT-encoded drug semantics, drug fingerprints, microbe genome sequence attributes, BERT-encoded microbe semantics, and microbe metabolic pathway attributes. We further introduce GNNExplainer into the graph convolutional network (GCN) pipeline to identify influential drug-microbe relations and feed the resulting structural signal back into graph refinement. In addition, we employ Elastic Weight Consolidation (EWC) to support parameter-efficient adaptation when biological data grow over time while mitigating catastrophic forgetting. Extensive experiments on three public datasets demonstrate that the proposed method achieves strong predictive performance, offers interpretable structural explanations, and provides a practical transfer-learning route for evolving MDA benchmarks. The source code for our model is publicly available at https://github.com/Luoxidu02/IDEAL.
SGTL-DDA is proposed, a novel graph transformer framework designed to incorporate structural information and domain-specific knowledge from heterogeneous biological information networks (HBINs) that successfully identifies both known therapeutics and novel repositioning candidates, supported by molecular docking results and literature evidence.
Bowei Zhao, Hui Zhao, Yu-an Huang et al.· IEEE transactions on computa...· 0 citations
The human microbiota is essential for maintaining physiological homeostasis, and microbial dysbiosis is increasingly implicated in the pathogenesis of complex diseases. Identifying potential microbe-disease associations (MDAs) can therefore facilitate mechanistic investigation, biomarker discovery, and therapeutic development. However, wet-laboratory validation is costly and time-consuming, while existing computational methods often struggle with sparse association networks and complex nonlinear interactions. We propose a novel deep learning approach named the Dual Dynamic Hypergraph Convolution Framework for Human Microbe-Disease Association Prediction (DDHMDA). Specifically, DDHMDA first utilizes graph convolutional networks to encode local topological features. Subsequently, it dynamically constructs a dual hypergraph architecture: a differentiable K-means similarity hypergraph to capture intra-modal global clustering patterns, and an attention-based cross-modal interaction hypergraph to model inter-modal interactions synergistically. Under leakage-free pair-level five-fold cross-validation (denoted as CV3), DDHMDA achieved AUC/AUPR values of 0.9789 ± 0.0177/0.9843 ± 0.0129 on HMDAD and 0.9651 ± 0.0042/0.9740 ± 0.0031 on Disbiome. DDHMDA also obtained the best overall CV3 performance among the eight evaluated methods. Furthermore, ablation experiments and case studies validate the practical effectiveness of individual modules and the biological interpretability in discovering novel MDAs. Therefore, DDHMDA would be a reliable tool for identifying potential MDAs.
Unexpected drug–drug interaction events (DDIEs) pose substantial clinical risks, yet many remain unannotated due to data scarcity and the rapid emergence of novel drug combinations. Conventional deep learning approaches struggle to generalize to these unseen interaction types and often lack interpretability under severe class imbalance. To address these challenges, we propose ZSCAN-DDIE, an interpretable zero-shot learning framework for DDIE prediction. The model integrates a biomedical pre-trained language model with an attention-based graph convolutional network (AGCN) to encode DDIE textual semantics and drug molecular structures, respectively. A cross-attention network (CAN) is introduced to align molecular substructures with pharmacological semantic components, enabling fine-grained cross-modal reasoning and improving interpretability at the substructure level. To mitigate modality bias and long-tailed distribution effects, we design a bimodal dynamic alignment (BDA) loss that combines hyperspherical embedding regularization with a stage-adaptive loss-switching mechanism. Experimental results under both conventional and generalized zero-shot settings demonstrate that ZSCAN-DDIE consistently outperforms state-of-the-art baselines across multiple evaluation metrics. The proposed framework not only enhances prediction accuracy for unseen DDIE categories but also provides biologically meaningful insights into molecular interaction mechanisms, offering a robust and clinically relevant solution for pharmacovigilance and drug safety assessment
The source code and data are available at https://github.com/GSX-0429/ZSCAN-DDIE.
Shaoxi Gao, Zhanpeng Gan, Fangfang Han et al.· Bioinformatics Advances· 0 citations
GCN-XGB is a novel hybrid computational framework that integrates a two-layer Graph Convolutional Network with Extreme Gradient Boosting to improve the accuracy of miRNA-disease association prediction and is suggested to be a powerful and reliable tool for identifying potential disease-related miRNAs and prioritizing candidates for experimental validation.
Jie Zhou, Peishen Yan, Jia Qu et al.· Journal of Mechanics in Medi...· 0 citations
DDI-MMAF is proposed, a lightweight cross-modal framework that avoids using explicit high-dimensional omics profiles as direct model inputs and enables accurate synergy prediction from raw minimalist inputs, offering a practical and efficient computational solution for cost-effective drug combination discovery.
Hao Li, Qianhui Jiang, Jiahui Guan et al.· European journal of medicina...· 0 citations
Microbe-drug association (MDA) prediction is of great importance for understanding drug action mechanisms and exploring microbiome-based therapeutic strategies. However, when confronted with extremely sparse biological networks with pronounced structural heterogeneity, existing methods often struggle to simultaneously model global topological dependencies and local structural disparities. To address these challenges, we propose MuSeL, a multi-scale adaptive graph representation learning framework that jointly mitigates the above issues from two complementary perspectives: global topology modeling and local structure optimization. Specifically, we develop a spectral kernel attention mechanism that leverages the normalized Laplacian and chebyshev polynomial expansion to efficiently capture multi-scale global topological relations in the spectral domain. Meanwhile, we construct a structure-aware self-adaptive sampling module that dynamically adjusts neighborhood sampling based on node clustering coefficients and degree centrality, thereby improving the reliability of local structural and feature representations. Finally, the fused multi-scale node embeddings are fed into a prediction module to estimate MDA scores. Extensive experiments show that outperforms existing mainstream models, while ablation studies and case analyses further validate its effectiveness and robustness. Overall, MuSeL provides a practical computational framework for prioritizing candidate MDAs, supporting downstream biological validation and drug repurposing research.
Yuehu Wu, Lei Wang, Zhengwei Li et al.· Proceedings of the 32nd ACM...· 0 citations