COCOA: A Framework for Fine-scale Mapping of Cell-type-specific Chromatin Compartments Using Epigenomic Information

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• Chromatin compartmentalization and epigenomic modifications are vital for cell differentiation and disease.
• High-resolution mapping of chromatin compartments typically requires extensive Hi-C or Micro-C sequencing data.
• Understanding the link between epigenomic modifications and compartment patterns is challenging.

Purpose of the Study:

• To develop a novel deep learning framework, COCOA, for inferring fine-scale chromatin compartment patterns.
• To leverage histone modification signals for predicting chromatin architecture.
• To provide a more accessible method for analyzing chromatin compartmentalization.

Main Methods:

• Utilized a deep neural network (COCOA) incorporating convolution and attention mechanisms.
• Processed epigenomic signals to extract 1D track features via bidirectional feature reconstruction.
• Integrated track features with contact features using attention and residual feature reduction to generate compartment patterns.

Main Results:

• COCOA accurately infers fine-scale chromatin compartment patterns from histone modification data.
• The framework demonstrates stable performance on test datasets.
• COCOA generates clearer and more detailed compartment patterns compared to high-depth experimental data at 1-kb resolution.
• In silico epigenomic perturbation experiments elucidated the impact of histone modifications on prediction.

Conclusions:

• COCOA offers a powerful tool for predicting chromatin compartmentalization using epigenomic data.
• The framework enables cell-type-specific prediction of chromatin patterns in various biological contexts.
• COCOA facilitates deeper understanding of chromatin organization in diverse biological scenarios and diseases.