A deep learning framework for comprehensive prediction of human RNA G-quadruplex-binding proteins

Area of Science:

• Computational Biology
• Molecular Biology
• Bioinformatics

Background:

• G-quadruplex-binding proteins (G4BPs) are vital for RNA metabolism and cellular stress responses.
• Experimental identification of G4BPs is challenging, necessitating computational approaches.

Purpose of the Study:

• To develop and validate a deep learning framework for predicting RNA G4BPs (RG4BPs).
• To identify novel RG4BP candidates within the human proteome.
• To provide a user-friendly web server for RG4BP prediction and analysis.

Main Methods:

• Integration of diverse encoding strategies and neural architectures, including ESM-2 protein language model embeddings and an LSTM architecture.
• Application of the trained model to the human proteome for candidate identification.
• Development of the G4REP web server for accessible RG4BP prediction.

Main Results:

• The best-performing deep learning model achieved 86% accuracy in identifying RG4BPs.
• Identified 2,160 high-confidence RG4BP candidates in the human proteome.
• Discovered that many RG4BP candidates possess intrinsically disordered regions and are enriched in stress granules, suggesting a link to cellular stress.

Conclusions:

• The study provides an effective computational approach to explore the landscape of RG4BPs.
• Novel RG4BP candidates and their potential roles in RNA regulation and stress response pathways were uncovered.
• The G4REP web server facilitates broad access to RG4BP prediction and analysis tools.