Related Experiment Video
Updated: Jun 13, 2026
03:37
Generating the Transcriptional Regulation View of Transcriptomic Features for Prediction Task and Dark Biomarker Detection on Small Datasets
Published on: March 1, 2024
OMNI: Optimized Multiview Network Integration with Heterogeneous Graph Attention for Biomedical Interaction
Gori Sankar Borah1, Sukriti Tiwari2, Selvaraman Nagamani1
1Advanced Computation and Data Sciences Division, CSIR-North East Institute of Science and Technology (NEIST), Jorhat, Assam 785006, India.
Journal of Chemical Information and Modeling
|June 11, 2026
Summary
This study introduces a novel multiview heterogeneous graph attention network (GAT) to accurately predict biomedical relationships, improving drug discovery and disease mechanism understanding. The model achieves superior performance in chemical-gene interaction prediction.
Area of Science:
- Biomedical Informatics
- Computational Biology
- Network Science
Background:
- Accurate prediction of biomedical relationships is crucial for disease mechanism understanding and drug discovery.
- Large-scale, heterogeneous biological data presents challenges for traditional network modeling approaches.
- Homogeneous graph neural networks (GNNs) and random walks struggle with high-order, diverse neighborhood information in complex biomedical graphs.
Purpose of the Study:
- To develop a scalable and efficient framework for predicting biomedical relationships using a multiview heterogeneous graph attention network (GAT).
- To capture comprehensive structural and semantic information from heterogeneous biomedical data.
- To outperform existing state-of-the-art methods in relation prediction tasks.
Main Methods:
- Applied a multiview heterogeneous graph attention network (GAT)-based architecture to aggregate heterogeneous interactions across multiple biomedical entity types.
- Utilized optimized aggregation strategies and multiprocessing for computationally efficient generation of high-quality node embeddings.
- Evaluated multiple decoder architectures, identifying a multilayer perceptron (MLP) as most effective for multitype relation classification.
Main Results:
- The proposed model achieved an AUROC of 0.90 for chemical-gene interaction prediction, outperforming state-of-the-art methods like CGINet, Node2Vec, HAN, HGT, and GCN-based BioNet.
- The model successfully identified top-ranking chemical-gene interactions in cancer and predicted gene-phytochemical relationships.
- The network comprises 124,604 unique nodes and 48,482,286 interactions, demonstrating scalability.
Conclusions:
- The developed GAT-based framework provides a scalable and powerful approach for biomedical relation prediction.
- This method significantly enhances the accuracy of predicting complex interactions within biomedical networks.
- The framework holds strong potential for applications in drug screening and uncovering disease mechanisms.
