PaRPI predicts RNA-Protein interactions from cross-protocol and cross-batch RNA-binding protein datasets.

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• RNA-binding proteins (RBPs) are crucial for gene expression regulation.
• Existing computational methods for predicting RNA-protein interactions are often limited by specific experimental protocols and data batches.
• A unified computational model is needed to capture diverse RBP-RNA interaction patterns.

Purpose of the Study:

• To develop PaRPI, a novel computational method for predicting RNA-protein binding sites.
• To create a unified model that integrates data from various experimental protocols and batches.
• To enable large-scale exploration of RNA-protein interactions for gene regulation and disease studies.

Main Methods:

• PaRPI employs a bidirectional RNA-protein selection approach.
• It groups RBP datasets by cell lines, integrating data from different protocols and batches.
• The method utilizes semantic embeddings to analyze interaction networks and disease variant impacts.

Main Results:

• PaRPI accurately identifies RNA-protein binding sites, outperforming state-of-the-art models on 261 datasets (eCLIP and CLIP-seq).
• The model demonstrates robust generalization, predicting interactions with novel RNA and protein receptors.
• PaRPI effectively analyzes disease variant impacts on RBP binding and dissects complex interaction networks.

Conclusions:

• PaRPI offers a powerful, unified computational approach for predicting RNA-protein binding sites.
• Its strong generalization capabilities facilitate the study of previously uncharacterized interactions.
• PaRPI advances research into gene regulation, RNA-protein interactions, and disease mechanisms.