preciseTAD: a transfer learning framework for 3D domain boundary prediction at base-pair resolution

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Chromosome conformation capture (Hi-C) reveals extensive DNA folding into 3D domains like Topologically Associating Domains (TADs) and chromatin loops.
• CTCF and cohesin binding at domain boundaries are crucial for maintaining 3D genome structure and function.
• Current methods can map 3D domains but struggle to define boundaries at the resolution of individual proteins.

Purpose of the Study:

• To develop a method for predicting genome 3D domain boundaries at base-pair resolution.
• To improve the understanding of how genomic regulators influence the 3D genome structure.
• To provide a tool that can delineate domain boundaries even without Hi-C data.

Main Methods:

• Developed preciseTAD, an optimized transfer learning framework.
• Trained the framework on high-resolution genome annotation data.
• Leveraged existing data to predict boundaries at base-pair resolution.

Main Results:

• preciseTAD accurately predicts genome 3D domain boundaries at base-pair resolution.
• Predicted boundaries show strong support from experimental evidence.
• The method successfully delineates boundaries in cells lacking Hi-C data.

Conclusions:

• preciseTAD offers a powerful framework for high-resolution 3D genome structure analysis.
• This approach enhances our understanding of genome organization and regulation.
• preciseTAD facilitates the study of 3D genome architecture at the protein-binding level.