Neighborhood Topology-Aware Knowledge Graph Learning and Microbial Preference Inferring for Drug-Microbe Association Prediction

Journal of Chemical Information and Modeling, Journal Year: 2025, Volume and Issue: 65(1), P. 435 - 445

Published: Jan. 2, 2025

The human microbiota may influence the effectiveness of drug therapy by activating or inactivating pharmacological properties drugs. Computational methods have demonstrated their ability to screen reliable microbe-drug associations and uncover mechanism which drugs exert functions. However, previous prediction failed completely exploit neighborhood topologies microbe entities diverse correlations between entity pair other entities. In addition, they ignored case that a prefers associate with its own specific A novel method, PCMDA, was proposed learning entities, inferring association preferences, integrating features each based on multiple biological premises. First, knowledge graph consisting microbe, disease, is established help subsequent integration topological structure similarity, interaction, relationship any two We generate various embeddings for (or drug) through random walks restarts microbe-disease-drug graph. Distance-level attention designed adaptively fuse covering ranges. Second, imply latent relationships while relational are derived from semantics connections among fused module multilayer perceptron networks. Third, considering preference tends especially group drugs, information-level integrate dependency microbial candidate drug. Finally, dual-gated network encode perspectives. comparative experiments seven state-of-the-art demonstrate PCMDA's superior performance prediction. studies three recall rate evaluation top-ranked candidates indicate PCMDA has capability discovering microbes associated datasets source codes freely available at https://github.com/pingxuan-hlju/pcmda.

Language: Английский

---

KNDM: A Knowledge Graph Transformer and Node Category Sensitive Contrastive Learning Model for Drug and Microbe Association Prediction

Journal of Chemical Information and Modeling, Journal Year: 2025, Volume and Issue: unknown

Published: April 23, 2025

It has been proven that the microbiome in human bodies can promote or inhibit treatment effects of drugs by affecting their toxicities and activities. Therefore, identifying drug-related microbes helps understanding how exert functions under influence these microbes. Most recent methods for microbe prediction are developed based on graph learning. However, those fail to fully utilize diverse characteristics drug entities from perspective a knowledge graph, as well contextual relationships among multiple meta-paths meta-path perspective. Moreover, previous overlook consistency between entity features derived node semantic extracted meta-paths. To address limitations, we propose knowledge-graph transformer category-sensitive contrastive learning-based association model (KNDM). This learns entities, encodes across meta-paths, integrates feature consistency. First, construct consisting which aids revealing similarities associations any two entities. Second, considering heterogeneity an integrate diversity types various them. Third, constructed capture embed nodes. A learning strategy with recursive gating is proposed specific individual while fusing Finally, develop node-category-sensitive enhance features. Extensive experiments demonstrate KNDM outperforms eight state-of-the-art drug-microbe models, ablation studies validate effectiveness its key innovations. Additionally, case candidate associated three drugs-curcumin, epigallocatechin gallate, ciprofloxacin-further showcase KNDM's capability identify potential associations.

Language: Английский

---

Boosting Drug-Disease Association Prediction for Drug Repositioning via Dual-Feature Extraction and Cross-Dual-Domain Decoding

Journal of Chemical Information and Modeling, Journal Year: 2025, Volume and Issue: unknown

Published: April 25, 2025

The extraction of biomedical data has significant academic and practical value in contemporary sciences. In recent years, drug repositioning, a cost-effective strategy for development by discovering new indications approved drugs, gained increasing attention. However, many existing repositioning methods focus on mining information from adjacent nodes networks without considering the potential inter-relationships between feature spaces drugs diseases. This can lead to inaccurate encoding, resulting biased mined drug-disease association information. To address this limitation, we propose model called Dual-Feature Drug Repurposing Neural Network (DFDRNN). DFDRNN allows two features (similarity association) encode A self-attention mechanism is utilized extract neighbor It incorporates dual-feature modules: single-domain (SDDFE) module extracting within single domain (drugs or diseases) cross-domain (CDDFE) across domains. By utilizing these modules, ensure more appropriate encoding cross-dual-domain decoder also designed predict associations both Our proposed outperforms six state-of-the-art four benchmark sets, achieving an average AUROC 0.946 AUPR 0.597. Case studies three diseases show that be applied real-world scenarios, demonstrating its repositioning.

Language: Английский

---

Drug repositioning by collaborative learning based on graph convolutional inductive network

Future Generation Computer Systems, Journal Year: 2024, Volume and Issue: 162, P. 107491 - 107491

Published: Aug. 22, 2024

Language: Английский

---

TarKG: A Comprehensive Biomedical Knowledge Graph for Target Discovery

Bioinformatics, Journal Year: 2024, Volume and Issue: 40(10)

Published: Oct. 1, 2024

Abstract Motivation Target discovery is a crucial step in drug development, as it directly affects the success rate of clinical trials. Knowledge graphs (KGs) offer unique advantages processing complex biological data and inferring new relationships. Existing biomedical KGs primarily focus on tasks such repositioning drug–target interactions, leaving gap construction tailored for target discovery. Results We established comprehensive KG focusing discovery, termed TarKG, by integrating seven existing KGs, nine public databases, traditional Chinese medicine knowledge databases. TarKG consists 1 143 313 entities 32 806 467 relations across 15 entity categories 171 relation types, all centered around 3 core types: Disease, Gene, Compound. provides specialized knowledges including chemical structures, protein sequences, or text descriptions. By using different embedding algorithms, we assessed completion capabilities particularly disease–target link prediction. In case studies, further examined TarKG’s ability to predict potential targets Alzheimer’s disease (AD) identify diseases potentially associated with metallo-deubiquitinase CSN5, literature analysis validation. Furthermore, provided user-friendly web server (https://tarkg.ddtmlab.org) that enables users perform retrieval inference TarKG. Availability implementation accessible at https://tarkg.ddtmlab.org.

Language: Английский

---