A deep learning model for characterizing protein-RNA interactions from sequences at single-base resolution.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Protein-RNA interactions are crucial for gene expression and RNA metabolism.
• Understanding these interactions is key to deciphering RNA dysregulation in diseases.
• Current methods for characterizing protein-RNA binding have limitations in resolution and motif discovery.

Purpose of the Study:

• To develop a deep learning model for predicting protein-RNA binding affinity from sequence data.
• To identify novel RNA-binding motifs and understand their functional relevance.
• To improve the prediction resolution of RNA-protein interactions and prioritize disease-associated mutations.

Main Methods:

• Developed Reformer, a deep learning model utilizing sequence data for binding affinity prediction.
• Trained Reformer on 225 enhanced cross-linking and immunoprecipitation sequencing (eCLIP-seq) datasets.
• Validated model predictions using electrophoretic mobility shift assays (EMSAs).

Main Results:

• Reformer achieved high accuracy in predicting protein-RNA binding affinity at single-base resolution.
• The model identified RNA-binding motifs not detectable by traditional eCLIP-seq methods.
• Learned motifs demonstrated a correlation with RNA processing functions.
• EMSAs confirmed Reformer's precision in quantifying mutation impacts on RNA regulation.

Conclusions:

• Reformer enhances the prediction resolution of RNA-protein interactions.
• The model aids in identifying functional RNA-binding motifs and prioritizing mutations affecting RNA regulation.
• This deep learning approach offers a powerful tool for studying RNA biology and disease mechanisms.