DNAcycP: a deep learning tool for DNA cyclizability prediction.

Area of Science:

• Genomics
• Biophysics
• Computational Biology

Background:

• DNA mechanical properties are crucial for DNA-dependent biological processes.
• The loop-seq assay enables high-throughput quantification of DNA fragment bendability.
• Understanding DNA intrinsic bendability is key to deciphering its functional roles.

Purpose of the Study:

• To develop a computational tool for predicting intrinsic DNA cyclizability.
• To leverage deep learning for accurate prediction of DNA bendability.
• To analyze conserved DNA mechanical properties across species and at regulatory elements.

Main Methods:

• Development of DNAcycP, a deep-learning software tool.
• Utilizing loop-seq experimental data for model training and validation.
• Comparative analysis of predicted DNA cyclizability (C-scores) with experimental data and genomic features.

Main Results:

• DNAcycP accurately predicts intrinsic DNA cyclizability with high fidelity.
• The C-score effectively distinguishes DNA fragments with varying loopability.
• Conserved high DNA bendability is observed at nucleosome dyads in yeast and mouse genomes.
• Significantly elevated DNA cyclizability is detected at CTCF binding sites in the mouse genome, a property conserved across mammals.

Conclusions:

• DNAcycP is a reliable tool for predicting DNA intrinsic cyclizability.
• DNA mechanical properties, specifically bendability, are conserved across species and associated with key genomic structures like nucleosomes and CTCF binding sites.
• The inherent mechanical properties of DNA, particularly at CTCF motifs, play a significant role in genome organization and function.