Integrating Single-Molecule Sequencing and Deep Learning to Predict Haplotype-Specific 3D Chromatin Organization in a...

Area of Science:

• Genomics
• Computational Biology
• Epigenetics

Background:

• The three-dimensional (3D) genome architecture is vital for gene regulation and human health.
• Short-read sequencing methods struggle to resolve complex genomic regions and individual haplotypes.
• Understanding 3D genome organization is crucial for deciphering gene regulation and disease mechanisms.

Purpose of the Study:

• To develop a deep learning model for accurate, haplotype-specific 3D genome organization prediction.
• To leverage long-read sequencing data (Fiber-seq) for comprehensive chromatin analysis.
• To investigate the impact of allelic X-inactivation and chromosomal translocations on genome structure.

Main Methods:

• Development of FiberFold, a deep learning model integrating convolutional neural networks and transformer architectures.
• Application of FiberFold to analyze multi-omic data from Fiber-seq long-read sequencing.
• Analysis of a cell line with allelic X-inactivation and a patient with a 13;X balanced translocation.

Main Results:

• FiberFold accurately predicts cell-type-specific and haplotype-specific 3D genome organization.
• Topologically Associated Domains (TADs) were found to be attenuated on the inactive X chromosome.
• Significant alterations in TADs were predicted around a 13;X translocation in a patient with a Mendelian disease.

Conclusions:

• FiberFold enables high-resolution analysis of 3D genome organization using long-read sequencing.
• The model provides insights into chromatin biology and the molecular basis of human diseases.
• Integrating long-read epigenomic data with deep learning offers powerful tools for genomic research.