PrismNet: predicting protein-RNA interaction using in vivo RNA structural information

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• RNA binding protein (RBP) and RNA interactions are crucial for post-transcriptional gene regulation.
• Existing RBP-RNA interaction prediction methods often rely on sequence-based RNA structure predictions, neglecting cellular environment variations.
• This limitation hinders the prediction of cell type-specific RBP-RNA interactions.

Purpose of the Study:

• To develop a novel computational tool, PrismNet, for predicting cell type-specific RBP-RNA interactions.
• To integrate in vivo RNA structural information with RBP binding data for enhanced prediction accuracy.
• To provide a web server for accessible prediction of RBP-RNA binding probabilities.

Main Methods:

• Utilized a deep learning model within the PrismNet web server.
• Integrated in vivo RNA secondary structures obtained from icSHAPE experiments.
• Incorporated RBP binding site information derived from UV cross-linking and immunoprecipitation (CLIP) data from the same cell lines.

Main Results:

• PrismNet successfully predicts cell type-specific RBP-RNA interactions by combining sequence and in vivo structure data.
• The 'Sequence & Structure' mode accepts RBP and RNA region information to output binding probability.
• Generated saliency maps and sequence-structure integrative motifs to elucidate binding mechanisms.

Conclusions:

• PrismNet offers a powerful deep learning approach to predict RBP-RNA interactions with cell type specificity.
• The integration of in vivo RNA structures significantly improves prediction accuracy compared to sequence-only methods.
• The freely available web server facilitates research in post-transcriptional regulation and RBP-RNA binding.