Predicting lncRNA-protein interactions with bipartite graph embedding and deep graph neural networks.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Long non-coding RNAs (lncRNAs) are vital in biological processes, and understanding their interactions with proteins is key to uncovering their functions.
• Computational methods are increasingly replacing traditional experiments for predicting lncRNA-protein associations.
• Existing computational approaches inadequately address the heterogeneity in lncRNA-protein interaction prediction and integrating this with graph neural networks remains challenging.

Purpose of the Study:

• To develop a novel computational model for accurately predicting lncRNA-protein interactions.
• To address the limitations of existing methods in handling the heterogeneity of lncRNA-protein association data.
• To improve the robustness and accuracy of lncRNA-protein interaction prediction using graph neural networks.

Main Methods:

• Constructed a deep architecture named BiHo-GNN, integrating homogeneous and heterogeneous network properties via bipartite graph embedding.
• Employed a data encoder for heterogeneous networks to capture molecular association mechanisms.
• Implemented a mutual optimization process between homogeneous and heterogeneous networks to enhance model robustness.

Main Results:

• Evaluated BiHo-GNN on four datasets for lncRNA-protein interaction prediction.
• Demonstrated that BiHo-GNN outperforms existing bipartite graph-based methods in prediction accuracy.
• Validated the model's effectiveness on benchmarking datasets.

Conclusions:

• BiHo-GNN successfully integrates bipartite and homogeneous graph networks for enhanced lncRNA-protein interaction prediction.
• The model accurately predicts and discovers potential lncRNA-protein associations.
• This approach offers a robust framework for understanding lncRNA molecular functions through interaction prediction.