NPI-GNN: Predicting ncRNA-protein interactions with deep graph neural networks

Area of Science:

• Computational biology
• Bioinformatics
• Molecular biology

Background:

• Noncoding RNAs (ncRNAs) are vital in biological processes.
• Experimental identification of ncRNA-protein interactions (NPIs) is resource-intensive.
• Computational methods offer efficient alternatives for NPI prediction.

Purpose of the Study:

• To evaluate existing machine learning methods for NPI prediction.
• To introduce a novel Graph Neural Network (GNN)-based method, NPI-GNN, for predicting NPIs.
• To assess the performance and robustness of NPI-GNN.

Main Methods:

• Collected and utilized five benchmarking datasets for NPI prediction.
• Evaluated and compared existing machine learning-based NPI prediction methods.
• Developed and applied an end-to-end GNN-based model (NPI-GNN) incorporating network and sequence information.

Main Results:

• NPI-GNN achieved performance comparable to state-of-the-art methods via 5-fold cross-validation.
• The method demonstrated capability in predicting novel NPIs using network and sequence data.
• NPI-GNN exhibited robustness, with minimal performance impact from reduced sequence information.

Conclusions:

• NPI-GNN represents the first end-to-end GNN predictor for NPIs.
• The developed method provides an efficient and robust computational approach for NPI prediction.
• Accessible datasets and source code facilitate further research and application.