diPaRIS: Dynamic and Interpretable Protein-RNA Interactions Prediction With U-Shaped Network and Novel Structure...

Area of Science:

• Molecular Biology
• Computational Biology
• Genomics

Background:

• Protein-RNA interactions are crucial for biological processes and disease.
• Existing computational methods struggle to capture nucleotide correlations within RNA structures.
• Accurate prediction of these interactions is essential for understanding gene function and disease.

Purpose of the Study:

• To develop a deep learning method, diPaRIS, for predicting dynamic protein-RNA interactions.
• To improve the accuracy and interpretability of protein-RNA interaction predictions.
• To integrate in vivo RNA structural information for enhanced predictive power.

Main Methods:

• Developed diPaRIS, a deep learning model utilizing a U-shaped network architecture.
• Introduced a novel encoding scheme for SHAPE-seq data to capture nucleotide correlations.
• Integrated in vivo RNA structures for a comprehensive representation.

Main Results:

• diPaRIS demonstrated superior performance across 44 datasets, achieving high accuracy, AUC, AUPR, and F1-scores.
• The model excelled in cross-cell line predictions, outperforming existing methods.
• Generated interpretable analyses, including sequence binding motifs and attribution maps.

Conclusions:

• diPaRIS accurately predicts dynamic protein-RNA interactions and enhances interpretability.
• The method provides insights into conserved binding patterns and functional interpretation of genetic variants.
• Findings facilitate understanding of gene-disease associations in complex diseases.