RPI-PLMGNN: Enhancing RNA-Protein Interaction Prediction with the Pretrained Large Language Models and Graph Neural...

Area of Science:

• Computational Biology
• Bioinformatics
• Molecular Biology

Background:

• RNA-protein interactions (RPIs) are crucial for gene regulation and disease.
• Traditional RPI prediction methods are slow and cannot leverage large language models.
• Current methods often rely on manual feature extraction, limiting their effectiveness.

Purpose of the Study:

• To develop an efficient and accurate method for predicting RNA-protein interactions.
• To overcome the limitations of traditional RPI prediction techniques.
• To integrate multimodal features and Graph Neural Networks for enhanced RPI prediction.

Main Methods:

• Proposed RPI-PLMGNN, a method combining multimodal feature fusion with a Graph Neural Networks framework.
• Utilized RNAErnie and ESM2 for RNA and protein sequence feature extraction.
• Extracted structural features using RNAFold and SOPMA, integrating them with sequence features.
• Employed a hybrid Graph Neural Network architecture with Graph Attention Network and Gated Graph Convolutional Network modules.

Main Results:

• RPI-PLMGNN achieved superior prediction performance across multiple benchmark datasets.
• Demonstrated excellent generalization ability with high accuracy in cross-species validation (up to 98.2%).
• The method effectively integrates sequence and structural information for accurate RPI prediction.

Conclusions:

• RPI-PLMGNN is an efficient and accurate tool for RNA-protein interaction prediction.
• The method offers significant advantages over traditional approaches by leveraging large language models and multimodal features.
• Provides a valuable resource for advancing the understanding of RPI mechanisms and facilitating RNA-targeted drug development.