G4Boost: a machine learning-based tool for quadruplex identification and stability prediction.

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• G-quadruplexes (G4s) are crucial secondary structures in guanine-rich nucleic acids.
• Accurate assessment of G4 structural stability is vital for understanding their biological roles.
• Existing methods require improvement for reliable G4 motif identification and stability prediction.

Purpose of the Study:

• To develop G4Boost, a novel decision tree-based tool for G4 motif identification.
• To predict the folding probability and thermodynamic stability of G4 structures.
• To provide an accurate, sequence-based method for G4 analysis.

Main Methods:

• Utilized a decision tree-based approach (G4Boost).
• Incorporated sequence intrinsic features, nucleotide composition, and estimated structural topologies.
• Trained and validated the model on known G4 structures.

Main Results:

• G4Boost achieved >93% accuracy and an F1-score of 0.96 for predicting quadruplex folding state.
• Predicted folding energy with high precision (RMSE of 4.28, R² of 0.95).
• Successfully validated on experimentally determined G4 structures from plants and humans.

Conclusions:

• G4Boost significantly outperforms existing machine-learning tools like DeepG4, Quadron, and G4RNA Screener.
• The tool offers high accuracy and F1-scores for G4 prediction.
• G4Boost is a valuable resource for studying gene regulation across diverse species.