Reconstructing spatial organizations of chromosomes through manifold learning

Area of Science:

• Genomics
• Computational Biology
• Biophysics

Background:

• Understanding chromosome spatial organization is key to eukaryotic gene regulation.
• Hi-C technology reveals genomic locus interaction frequencies, offering insights into 3D genome folding.
• Existing methods often rely on assumptions linking interaction frequencies to spatial distances.

Purpose of the Study:

• To develop a novel computational framework, GEM (Genomic organization reconstructor based on conformational Energy and Manifold learning), for reconstructing 3D chromosome structures.
• To integrate Hi-C data with biophysical principles for more accurate genome modeling.
• To overcome limitations of previous methods by directly embedding Hi-C affinities into 3D space.

Main Methods:

• Developed GEM, a manifold learning-based framework for 3D genome reconstruction.
• Integrated Hi-C interaction data with biophysical feasibility constraints.
• Employed a novel approach to embed neighboring affinities from Hi-C space directly into 3D Euclidean space.

Main Results:

• GEM significantly outperformed existing state-of-the-art modeling methods in reconstructing 3D chromosome organization.
• The reconstructed 3D chromatin structures generated by GEM were validated as physically and physiologically sound.
• Applied GEM to successfully recover long-range genomic interactions previously missing in Hi-C data.

Conclusions:

• GEM offers a powerful and accurate method for reconstructing 3D genome structures by integrating Hi-C data and manifold learning.
• The framework provides biologically relevant 3D representations of chromosome organization.
• GEM's ability to recover missing genomic interactions opens new avenues for analyzing genome topology.