DeepZ: A Deep Learning Approach for Z-DNA Prediction.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Z-DNA is a non-canonical DNA structure with potential regulatory roles.
• Accurate identification of Z-DNA regions across the genome is challenging.
• Integrating diverse biological data can improve Z-DNA prediction.

Purpose of the Study:

• To develop a deep learning approach for whole-genome Z-DNA annotation.
• To identify key sequence and epigenetic features determining functional Z-DNA regions.
• To leverage multi-modal data for enhanced Z-DNA detection.

Main Methods:

• Utilized deep learning neural networks, including Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN).
• Aggregated information from DNA sequence, nucleotide properties (physical, chemical, structural), and omics data (histone modifications, methylation, chromatin accessibility, transcription factor binding).
• Incorporated data from Next-Generation Sequencing (NGS) experiments.

Main Results:

• Successfully trained a model for whole-genome Z-DNA region annotation.
• Performed feature importance analysis to identify critical determinants for functional Z-DNA.
• Demonstrated the capability to predict Z-DNA with high accuracy using integrated data.

Conclusions:

• The developed deep learning approach enables accurate genome-wide Z-DNA annotation.
• Key sequence and epigenetic features influencing Z-DNA formation and function have been identified.
• This method provides a powerful tool for studying Z-DNA's role in biological processes.