Examining the dynamics of three-dimensional genome organization with multitask matrix factorization.

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Three-dimensional (3D) genome organization is crucial for gene regulation.
• High-throughput Hi-C data allows examination of 3D genome structure changes.
• Detecting higher-order structural changes across multiple datasets is challenging.

Purpose of the Study:

• To develop a computational method for systematic detection of higher-order structural changes in 3D genome organization across multiple conditions.
• To address limitations of existing methods in modeling higher-order structures and dynamics across more than two conditions.

Main Methods:

• Tree-guided integrated factorization (TGIF), a multitask nonnegative matrix factorization (NMF) approach.
• TGIF is applicable to time series or hierarchically related biological conditions.
• Benchmarking on simulated and real Hi-C data.

Main Results:

• TGIF identifies large-scale changes (compartment/subcompartment) and local changes (TAD boundaries).
• TGIF boundaries demonstrate improved accuracy, reproducibility, and CTCF enrichment.
• TGIF detects differential regions associated with regulatory signals, gene expression, and tissue-specific processes.

Conclusions:

• TGIF is a flexible tool for analyzing 3D genome organization dynamics.
• It enables the examination of changes across disease and developmental processes.
• TGIF can prioritize sequence variants for phenotypes using GWAS catalog data.