G4detector: Convolutional Neural Network to Predict DNA G-Quadruplexes.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• G-quadruplexes (G4s) are nucleic acid structures in guanine-rich regions.
• G4 formation impacts gene regulation, chromatin, and is linked to diseases like cancer.
• Experimental G4 detection (G4-seq) is resource-intensive for whole genomes.

Purpose of the Study:

• Develop a computational method for accurate G4 prediction in DNA sequences.
• Improve G4 prediction by integrating sequence and RNA secondary structure information.
• Provide a scalable tool for genome-wide G4 analysis.

Main Methods:

• Developed G4detector, a convolutional neural network (CNN) model.
• Utilized G4-seq data across multiple species for training and validation.
• Incorporated both DNA sequence and RNA secondary structure data.

Main Results:

• G4detector significantly outperforms existing methods on benchmark datasets.
• Achieved high accuracy in genome-wide G4 detection.
• Demonstrated successful extrapolation of human-trained models to non-human species.

Conclusions:

• G4detector offers a powerful and accurate computational approach for G4 prediction.
• The method facilitates large-scale genomic studies of G-quadruplexes.
• Publicly available code and benchmarks enable broader research application.