DeepNanoHi-C: deep learning enables accurate single-cell nanopore long-read data analysis and 3D genome...

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell long-read concatemer sequencing (scNanoHi-C) offers insights into 3D genome organization.
• Existing analytical tools are insufficient for scNanoHi-C data challenges like sparsity and cell variability.

Purpose of the Study:

• To introduce DeepNanoHi-C, a deep learning framework tailored for scNanoHi-C data analysis.
• To accurately predict chromatin interactions and capture cell-specific 3D genome structures.

Main Methods:

• Utilizes a multistep autoencoder to capture global chromatin contact patterns.
• Employs a Sparse Gated Mixture of Experts (SGMoE) for dynamic expert selection based on chromatin patterns.
• Integrates multiscale predictions via a dual-channel prediction net for refined interaction information.

Main Results:

• DeepNanoHi-C outperforms existing methods in cell type distinction and data imputation.
• Successfully identifies cell-specific topologically associating domain (TAD) boundaries.
• Reveals conserved genomic structures across species, indicating evolutionary conservation of chromatin organization.

Conclusions:

• DeepNanoHi-C provides a robust framework for analyzing scNanoHi-C data, improving the understanding of 3D genome organization.
• The tool accurately models complex chromatin interactions and cell-specific features.
• Offers potential for broader applications in comparative genomics and evolutionary studies.