Machine learning-based prediction of DNA G-quadruplex folding topology with G4ShapePredictor.

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Deoxyribonucleic acid (DNA) forms complex four-stranded structures called G-quadruplexes (G4s) with varied folding patterns.
• G4 topologies (parallel, antiparallel, hybrid) are classified by strand orientation and are vital in DNA replication, repair, and transcription.
• G4s are significant in drug design, but predicting their folding topology from sequence remains a challenge.

Purpose of the Study:

• To introduce G4ShapePredictor (G4SP), a novel computational tool for predicting G4 folding topologies based on DNA sequence.
• To provide an accurate method for classifying G4 structures into parallel (4+0), antiparallel (2+2), or hybrid (3+1) topologies.

Main Methods:

• Development of G4ShapePredictor (G4SP), utilizing multi-classification machine learning models.
• Training G4SP on a custom dataset combining literature data and experimental circular dichroism results.
• Incorporation of a threshold optimization strategy to enhance prediction precision.

Main Results:

• G4ShapePredictor accurately predicts G4 folding topologies in potassium buffer from primary DNA sequences.
• The tool successfully classifies G4 structures into parallel (4+0), antiparallel (2+2), and hybrid (3+1) types.
• Identification of three sequence motifs associated with specific G4 folding topologies.

Conclusions:

• G4ShapePredictor offers a significant advancement in predicting G4 folding topology from sequence data.
• This tool can aid researchers in understanding G4 structure-function relationships and in designing G4-targeting drugs.
• The identified sequence motifs provide further insights into G4 structural determinants.