TL;DR
This paper introduces Dip3D, a novel computational framework that leverages long-read Pore-C sequencing data to reconstruct high-resolution diploid 3D human genomes. Dip3D overcomes key challenges in analyzing noisy Pore-C data, enabling unprecedented insights into haplotype-specific genome organization and its link to gene expression.
What Problem Does This Paper Address?
Understanding the 3D organization of the genome within the nucleus is crucial for deciphering fundamental biological processes like gene regulation, DNA replication, and cell development. In diploid organisms, the spatial arrangement of homologous chromosomes is particularly important, as it can influence gene expression and other epigenetic modifications. While technologies like Hi-C have enabled mapping of these 3D structures, reconstructing diploid genomes with high resolution has been challenging. This is due to limitations in Hi-C data, such as short read lengths and difficulty in assigning fragments to specific haplotypes (maternal vs. paternal). The advent of long-read sequencing technologies like Oxford Nanopore's Pore-C offers the potential for longer fragments and higher-order interactions, but also introduces its own set of challenges, including higher sequencing error rates and complex data analysis. This paper aims to bridge this gap by developing a robust method, Dip3D, to reconstruct high-resolution diploid 3D genomes using Pore-C data alone.
Methods
Dip3D is a seven-step pipeline designed to process Pore-C data for diploid 3D genome reconstruction:
- Read Mapping: Pore-C reads are decomposed into fragments and mapped to the reference genome using Falign, preserving fragment order.
- Genome-wide SNV Calling: A retrained deep learning model (Clair3) specifically optimized for Pore-C data is used to identify single nucleotide variants (SNVs).
- SNV Phasing: The identified SNVs are phased (assigned to maternal or paternal haplotypes) using a pipeline that converts Pore-C reads into pairwise contacts and utilizes HapCUT2.
- Fragment Haplo-tagging: Pore-C fragments are assigned to a specific haplotype using optimized filters based on mapping quality, alignment identity, and SNV context.
- Haplotype Imputation: For fragments that could not be directly haplo-tagged, an imputation strategy is employed, leveraging distance rules, bridging rules, and dominant haplotype rules.
- Formatted Output: The processed data is output in a specialized SAM format.
- Applications: Supplementary tools are provided for extracting diploid contact matrices and analyzing haplotype-specific high-order interactions.
The key innovations lie in the specialized Pore-C SNV calling model, the optimized haplo-tagging filters to reduce inter-haplotype (h-trans) contacts, and the multi-step haplotype imputation strategy that effectively utilizes the long reads.
Datasets
- Human Cell Lines: HG001 and HG002 (NA12878 and NA24385, respectively) were used for training and evaluation.
- F1 Hybrid Mice: Generated from a C57BL/6J × PWK/PhJ cross, these mice were used to study haplotype constitution due to their higher heterozygous SNV density.
- Benchmark Datasets:
- GIAB v3.3.2 truth variants for HG001 and HG002 were used for SNV calling evaluation.
- Hi-C data from Rao et al. (2014) for HG001 and from Du et al. (2017) for F1 mice were used for comparative analysis.
Key Insights and Ideas
- Pore-C's Advantage for Diploid Genomes: The long read lengths and ability to capture high-order interactions in Pore-C data provide a distinct advantage over Hi-C for diploid genome analysis, mitigating limitations like short read length and GC bias.
- Haplotype Imputation is Crucial: While Pore-C offers a higher haplo-tagging rate than Hi-C, direct haplo-tagging is still limited. Dip3D's imputation strategy is essential for maximizing data utilization and achieving high-resolution diploid structures.
- Haplotype-Specific Interactions Drive Biological Processes: The study reveals that haplotype-specific high-order interactions play a significant role in processes like X-chromosome inactivation (XCI) and are directly involved in allele-specific expression (ASE) related to XCI.
- Pore-C Uncovers Hidden Structures: Dip3D's ability to capture more comprehensive and continuous genomic structures, particularly in "difficult" genomic regions, allows for the discovery of TADs and other structures that were previously obscured in Hi-C data.
Key Results
- Improved SNV Calling: The retrained Pore-C model achieved high recall rates (up to 98.6%) and precision (≥99%) for heterozygous SNVs, outperforming the original Clair3 model.
- Reduced h-trans Rate: Dip3D's haplo-tagging strategy reduced the h-trans contact rate from 10.1% to 5.6%, critical for accurate diploid genome reconstruction.
- Enhanced Haplo-tagging and Imputation: The imputation strategy increased the haplotype-assigned contact rate by 11.9-fold to 72.7% on HG001 data.
- Superior Resolution and Completeness: Dip3D achieved a 10-fold higher resolution (5 kb vs. 50 kb) and 6.9-fold higher haplotype-assigned contact density compared to Hi-C. It also significantly reduced gap regions (9.4% vs. 73.3%).
- More TADs Detected: Dip3D reproduced 2.3 times more TAD structures than Hi-C, particularly in regions with sparse Hi-C coverage.
- Link to Allele-Specific Expression: The study found that haplotype-specific high-order interactions are directly involved in ASE related to XCI, while genetic imprinting on autosomes appears to be regulated by other mechanisms.
Limitations and Future Work
- Pore-C Sequencing Errors: While Dip3D addresses challenges, Pore-C still inherits higher sequencing error rates and uneven error distribution from nanopore technology, which can complicate SNV identification and haplo-tagging.
- Haplotype Imputation Accuracy: While effective, haplotype imputation still relies on inference, and the accuracy can be influenced by the quality and density of phased SNVs.
- Future Work: The authors suggest that Dip3D's ability to reconstruct high-resolution diploid 3D genomes will enable further investigation into biological processes associated with 3D genome structure. Given the ongoing improvements in nanopore sequencing, Pore-C is poised to become an increasingly important tool for 3D genome studies, with Dip3D serving as a key analytical framework.
Summary updated