Predicting dynamic cellular protein-RNA interactions by deep learning using in vivo RNA structures.

Area of Science:

• Molecular Biology
• Computational Biology
• Genomics

Background:

• RNA-binding proteins (RBPs) interactions are vital for RNA function and cellular regulation.
• Current prediction tools lack the ability to capture condition-dependent RBP-RNA interactions.
• Understanding dynamic RBP-RNA binding is essential for cellular processes and disease research.

Purpose of the Study:

• To develop a novel deep learning tool, PrismNet, for predicting dynamic RNA-binding protein (RBP)-RNA interactions.
• To integrate experimental in vivo RNA secondary structure data with RBP binding data for accurate prediction.
• To capture the condition-dependent nature of RBP-RNA interactions.

Main Methods:

• Profiling transcriptome-wide in vivo RNA secondary structures across seven cell types.
• Developing PrismNet, a deep learning model integrating RNA structure and RBP binding data.
• Utilizing an attention mechanism within PrismNet to identify precise RBP binding sites.

Main Results:

• PrismNet accurately predicts dynamic RBP binding across various cellular conditions for 168 RBPs.
• The tool enhances understanding of CLIP-seq data and extends interaction data to new cell types.
• Identified enrichment of structure-changing variants (riboSNitches) at dynamic RBP binding sites.

Conclusions:

• PrismNet provides access to cell-type-specific RBP-RNA interactions, previously inaccessible.
• The findings link genetic diseases to dysregulated RBP binding through structure-changing variants.
• PrismNet offers significant utility for understanding and potentially treating human diseases.