IRGL-RRI: interpretable graph representation learning for plant RNA-RNA interaction discovery

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• Plant RNAs play vital roles in gene expression and protein synthesis.
• Understanding RNA-RNA interactions (RRIs) is crucial but challenging due to their complexity.
• Current deep learning methods for RRI prediction lack sufficient accuracy.

Purpose of the Study:

• To develop an interpretable graph representation model for accurate plant RRI prediction.
• To enhance the efficiency and accuracy of predicting plant RNA-RNA interactions.
• To improve the interpretability of RRI prediction models.

Main Methods:

• Proposed an interpretable graph representation model for plant RRI prediction.
• Enriched RNA data by extracting base features and reconstructing them hierarchically.
• Employed a masking strategy and regularization for enhanced RNA feature extraction.
• Combined Kolmogorov-Arnold Networks (KAN) and multi-scale fusion for RRI modeling and interpretability.

Main Results:

• The model accurately identifies potential plant RNA-RNA interactions (RRIs).
• Demonstrated improved prediction accuracy compared to existing deep learning methods.
• Case studies confirmed the model's effectiveness on public datasets.

Conclusions:

• The proposed interpretable graph model is a powerful tool for plant RRI prediction.
• This approach has significant potential for plant gene function annotation.
• The model offers enhanced accuracy and interpretability in predicting complex RNA interactions.