HINGRL: predicting drug–disease associations with graph representation learning on heterogeneous information networks

Briefings in Bioinformatics, Journal Year: 2021, Volume and Issue: 23(1)

Published: Nov. 16, 2021

Abstract Identifying new indications for drugs plays an essential role at many phases of drug research and development. Computational methods are regarded as effective way to associate with indications. However, most them complete their tasks by constructing a variety heterogeneous networks without considering the biological knowledge diseases, which believed be useful improving accuracy repositioning. To this end, novel information network (HIN) based model, namely HINGRL, is proposed precisely identify on graph representation learning techniques. More specifically, HINGRL first constructs HIN integrating drug–disease, drug–protein protein–disease diseases. Then, different strategies applied learn features nodes in from topological perspectives. Finally, adopts Random Forest classifier predict unknown drug–disease associations integrated diseases obtained previous step. Experimental results demonstrate that achieves best performance two real datasets when compared state-of-the-art models. Besides, our case studies indicate simultaneous consideration topology allows more comprehensive perspective. The promising also reveals utilization rich provides alternative view especially

Language: Английский

---

Predicting miRNA–disease associations via learning multimodal networks and fusing mixed neighborhood information

Briefings in Bioinformatics, Journal Year: 2022, Volume and Issue: 23(5)

Published: April 14, 2022

Abstract Motivation In recent years, a large number of biological experiments have strongly shown that miRNAs play an important role in understanding disease pathogenesis. The discovery miRNA–disease associations is beneficial for diagnosis and treatment. Since inferring these through time-consuming expensive, researchers sought to identify the utilizing computational approaches. Graph Convolutional Networks (GCNs), which exhibit excellent performance link prediction problems, been successfully used association prediction. However, GCNs only consider 1st-order neighborhood information at one layer but fail capture from high-order neighbors learn miRNA representations propagation. Therefore, how aggregate effectively explicit way still challenging. Results To address such challenge, we propose novel method called mixed (MINIMDA), could fuse diseases multimodal networks. First, MINIMDA constructs integrated similarity network respectively with their multisource information. Then, embedding are obtained by fusing network, Finally, concentrate feed them into multilayer perceptron (MLP) predict underlying associations. Extensive experimental results show superior other state-of-the-art methods overall. Moreover, outstanding on case studies esophageal cancer, colon tumor lung cancer further demonstrates effectiveness MINIMDA. Availability implementation https://github.com/chengxu123/MINIMDA http://120.79.173.96/

Language: Английский

---

Hierarchical graph attention network for miRNA-disease association prediction

Molecular Therapy, Journal Year: 2022, Volume and Issue: 30(4), P. 1775 - 1786

Published: Feb. 1, 2022

Many biological studies show that the mutation and abnormal expression of microRNAs (miRNAs) could cause a variety diseases. As an important biomarker for disease diagnosis, miRNA is helpful to understand pathogenesis, promote identification, diagnosis treatment However, pathogenic mechanism how miRNAs affect these diseases has not been fully understood. Therefore, predicting potential miRNA-disease associations great importance development clinical medicine drug research. In this study, we proposed novel deep learning model based on hierarchical graph attention network (HGANMDA). Firstly, constructed miRNA-disease-lncRNA heterogeneous known associations, miRNA-lncRNA disease-lncRNA associations. Secondly, node-layer was applied learn neighbor nodes different meta-paths. Thirdly, semantic-layer Finally, bilinear decoder employed reconstruct connections between The extensive experimental results indicated our achieved good performance satisfactory in

Language: Английский

---

Likelihood-based feature representation learning combined with neighborhood information for predicting circRNA–miRNA associations

Briefings in Bioinformatics, Journal Year: 2024, Volume and Issue: 25(2)

Published: Jan. 22, 2024

Connections between circular RNAs (circRNAs) and microRNAs (miRNAs) assume a pivotal position in the onset, evolution, diagnosis treatment of diseases tumors. Selecting most potential circRNA-related miRNAs taking advantage them as biological markers or drug targets could be conducive to dealing with complex human through preventive strategies, diagnostic procedures therapeutic approaches. Compared traditional experiments, leveraging computational models integrate diverse data order infer associations proves more efficient cost-effective approach. This paper developed model Convolutional Autoencoder for CircRNA-MiRNA Associations (CA-CMA) prediction. Initially, this merged natural language characteristics circRNA miRNA sequence features circRNA-miRNA interactions. Subsequently, it utilized all pairs construct molecular association network, which was then fine-tuned by labeled samples optimize network parameters. Finally, prediction outcome is obtained utilizing deep neural networks classifier. innovatively combines likelihood objective that preserves neighborhood optimization, learn continuous feature representation words preserve spatial information two-dimensional signals. During process 5-fold cross-validation, CA-CMA exhibited exceptional performance compared numerous prior approaches, evidenced its mean area under receiver operating characteristic curve 0.9138 minimal SD 0.0024. Furthermore, recent literature has confirmed accuracy 25 out top 30 identified highest scores during case studies. The results these experiments highlight robustness versatility our model.

Language: Английский

---

lncRNA-disease association prediction based on optimizing measures of multi-graph regularized matrix factorization

Computer Methods in Biomechanics & Biomedical Engineering, Journal Year: 2025, Volume and Issue: unknown, P. 1 - 16

Published: March 20, 2025

In this paper, we propose a novel lncRNA-disease association prediction algorithm based on optimizing measures of multi-graph regularized matrix factorization (OM-MGRMF). The method first calculates the semantic similarity diseases, functional lncRNAs, and Gaussian both. It then constructs new by using K-nearest-neighbor (KNN) algorithm. Finally, objective function is constructed through utilization ranking regularization constraints. This iteratively optimized an adaptive gradient descent experimental results OM-MGRMF outperform those classical methods in both K-fold cross-validation.

Language: Английский

---

SPRDA: a link prediction approach based on the structural perturbation to infer disease-associated Piwi-interacting RNAs

Briefings in Bioinformatics, Journal Year: 2022, Volume and Issue: 24(1)

Published: Nov. 9, 2022

piRNA and PIWI proteins have been confirmed for disease diagnosis treatment as novel biomarkers due to its abnormal expression in various cancers. However, the current research is not strong enough further clarify functions of cancer underlying mechanism. Therefore, how provide large-scale serious candidates biological has grown up be a pressing issue. In this study, computational model based on structural perturbation method proposed predict potential disease-associated piRNAs, called SPRDA. Notably, SPRDA belongs positive-unlabeled learning, which unaffected by negative examples contrast previous approaches. 5-fold cross-validation, shows high performance benchmark dataset piRDisease, with an AUC 0.9529. Furthermore, predictive 10 diseases robustness method. Overall, approach can unique insights into pathogenesis will advance field oncology treatment.

Language: Английский

---

A machine learning framework based on multi-source feature fusion for circRNA-disease association prediction

Briefings in Bioinformatics, Journal Year: 2022, Volume and Issue: 23(5)

Published: Sept. 1, 2022

Abstract Circular RNAs (circRNAs) are involved in the regulatory mechanisms of multiple complex diseases, and identification their associations is critical to diagnosis treatment diseases. In recent years, many computational methods have been designed predict circRNA-disease associations. However, most existing rely on single correlation data. Here, we propose a machine learning framework for association prediction, called MLCDA, which effectively fuses sources heterogeneous information including circRNA sequences disease ontology. Comprehensive evaluation gold standard dataset showed that MLCDA can successfully capture relationships between circRNAs diseases accurately potential addition, results case studies real data show significantly outperforms other methods. serve as useful tool providing mechanistic insights research thus facilitating progress treatment.

Language: Английский

---

Artificial intelligence accelerates multi-modal biomedical process: A Survey

Neurocomputing, Journal Year: 2023, Volume and Issue: 558, P. 126720 - 126720

Published: Aug. 22, 2023

Language: Английский

---

AMDECDA: Attention Mechanism Combined With Data Ensemble Strategy for Predicting CircRNA-Disease Association

IEEE Transactions on Big Data, Journal Year: 2023, Volume and Issue: 10(4), P. 320 - 329

Published: Nov. 20, 2023

Accumulating evidence from recent research reveals that circRNA is tightly bound to human complex disease and plays an important regulatory role in progression. Identifying disease-associated occupies a key the of pathogenesis. In this study, we propose new model AMDECDA for predicting circRNA-disease association (CDA) by combining attention mechanism data ensemble strategy. Firstly, fuse heterogeneous information including Gaussian interaction profile (GIP), semantics GIP, then use Graph Attention Network (GAT) focus on critical data, reasonably allocate resources extract their essential features. Finally, deep RVFL network (edRVFL) utilized quickly accurately predict CDA non-iterative manner closed-form solutions. five-fold cross-validation experiment benchmark set, achieves accuracy 93.10% with sensitivity 97.56% 0.9235 AUC. comparison previous models, exhibits highly competitiveness. Furthermore, 26 top 30 unknown CDAs predicted scores are proved related literature. These results indicate can effectively anticipate latent provide help further biological wet experiments.

Language: Английский

---

Predicting RNA structures and functions by artificial intelligence

Trends in Genetics, Journal Year: 2023, Volume and Issue: 40(1), P. 94 - 107

Published: Oct. 26, 2023

Language: Английский

---