Interpretable deep residual network uncovers nucleosome positioning and associated features.

Area of Science:

• Genomics
• Epigenetics
• Computational Biology

Background:

• Nucleosomes, the basic units of eukaryotic chromosomes, are crucial for gene regulation and DNA-related processes.
• Genomic positioning of nucleosomes is influenced by DNA sequences and impacts gene expression, replication, methylation, and repair.
• Understanding nucleosome organization is key to deciphering epigenetic regulation.

Purpose of the Study:

• To develop an interpretable deep learning framework (NuPoSe) for identifying nucleosome positions and associated features genome-wide.
• To analyze sequence and structural patterns influencing nucleosome organization in the human genome.
• To assess the predictive power of identified features, including those from linker DNA.

Main Methods:

• Development of NuPoSe, a deep residual network framework.
• Training NuPoSe on high-coverage human MNase-seq data.
• Identification and analysis of 43 informative sequence and structural features related to nucleosome positioning.

Main Results:

• NuPoSe effectively learns sequence and structural patterns of nucleosome organization.
• Identified features, primarily tri- and di-nucleotides, are significantly associated with nucleosomal DNA periodicity and histone octamer location.
• Linker DNA features contribute up to 10% to the prediction model's accuracy.
• NuPoSe achieves 80-89% classification accuracy on independent datasets across different organisms and cell types.

Conclusions:

• NuPoSe provides an accurate and interpretable method for predicting nucleosome positions genome-wide.
• Sequence and structural features, including those in linker DNA, are critical determinants of nucleosome organization.
• The framework's applicability to diverse datasets highlights its potential for broad use in epigenetics and genomics research.