Direct detection of alternative DNA conformations with long-read sequencing and machine learning approaches

Area of Science:

• Molecular Biology
• Genomics
• Bioinformatics

Background:

• Identifying non-canonical DNA structures like G-quadruplexes (G4s) in live cells is crucial but limited by low-resolution and GC-biased short-read sequencing.
• Long-read sequencing technologies offer potential solutions to overcome current methodological constraints in detecting these structures.

Purpose of the Study:

• To investigate the utility of Oxford Nanopore Technologies (ONT) long-read sequencing for detecting G4 and other non-B DNA structures in vitro and in live cells.
• To apply machine learning algorithms to ONT sequencing data for accurate identification of G4 DNA conformations.

Main Methods:

• Applied ONT sequencing to synthetic oligos forming G4s and DNA from live cells, probed with potassium permanganate to detect single-stranded DNA.
• Utilized machine learning models (logistic regression, random forest, XGBoost, 1D CNNs) to analyze raw ONT sequencing current data.
• Sequenced permanganate-modified DNA directly using ONT to assess effects on sequencing metrics.

Main Results:

• ONT sequencing successfully resolved G4 structures, with error profiles dependent on the G4's in vitro structural state.
• Machine learning models achieved ~90% accuracy in distinguishing G4 motifs in G4 vs. B-form conformations from ONT data.
• Direct sequencing of permanganate-modified DNA using ONT showed minimal impact on read length, yield, and alignment accuracy.

Conclusions:

• ONT long-read sequencing, coupled with machine learning, is a viable method for detecting G4 and other non-B DNA structures in cells.
• High-depth ONT sequencing can differentiate B-folded from single-stranded DNA regions, revealing frequent ssDNA presence in non-B DNA structures.