Predicting CTCF-mediated chromatin interactions by integrating genomic and epigenomic features

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• The CCCTC-binding protein (CTCF) network mediates long-range chromatin interactions vital for genome organization and function.
• This network was previously thought to be invariant, but recent findings suggest cell-type-specific interactions play a role in cell identity.

Purpose of the Study:

• To develop a machine learning framework, Lollipop, for predicting CTCF-mediated long-range chromatin interactions.
• To identify genomic and epigenomic features that determine CTCF-mediated chromatin interactions and their impact on gene expression.

Main Methods:

• Developed Lollipop, a machine learning framework utilizing genomic and epigenomic data.
• Validated predictions using ChIA-PET data as a benchmark.
• Confirmed predictions computationally and experimentally via Chromatin Conformation Capture (3C).

Main Results:

• Lollipop accurately predicts CTCF-mediated chromatin interactions within and across cell types.
• The framework outperforms methods relying solely on CTCF motif orientation.
• Identified gene expression within loops as an additional determinant of CTCF-mediated chromatin wiring.

Conclusions:

• The CTCF network exhibits significant cell-type-specific interactions contributing to cell identity.
• Lollipop provides an accurate and robust method for predicting these interactions.
• The study enhances understanding of the principles governing CTCF-mediated chromatin interactions and their regulatory role in gene expression.