Deciphering the 3D genome organization across species from Hi-C data

Area of Science:

• Genomics
• Computational Biology
• Evolutionary Biology

Background:

• 3D genome organization is crucial for gene regulation but varies across species due to unique factors and DNA sequences.
• Understanding species-specific chromatin folding mechanisms is challenging for cross-species comparisons.

Purpose of the Study:

• To develop a machine learning model, Chimaera, for predicting 3D genome organization (Hi-C maps) from DNA sequences.
• To explore genome folding patterns across evolution and identify sequence elements influencing chromatin structure.
• To investigate evolutionary relationships based on chromatin organization.

Main Methods:

• Utilized Hi-C data and convolutional neural networks (CNNs) to train Chimaera.
• Employed Chimaera's latent representations for unsupervised discovery of chromatin features and structural signatures.
• Applied Chimaera across multiple species, including vertebrates, Drosophila melanogaster, and Dictyostelium discoideum, for cross-species analysis.

Main Results:

• Chimaera successfully predicted Hi-C maps and revealed an atlas of chromatin features like insulation and loops.
• Identified conserved insulator roles of CTCF (vertebrates) and BEAF-32 (Drosophila), and discovered a new insulator motif in Drosophila.
• Demonstrated that gene orientation influences loop formation in Dictyostelium and linked chromatin folding to gene location in other organisms.
• Cross-species predictions facilitated the construction of a chromatin structure-based evolutionary tree from plants to mammals.

Conclusions:

• Chimaera provides a powerful tool for exploring 3D genome organization and its evolutionary dynamics.
• The study highlights the interplay between DNA sequence, gene location, and chromatin folding across diverse taxa.
• Revealed conserved and novel mechanisms of genome organization, advancing our understanding of evolutionary relationships through chromatin structure.