The application of high-throughput chromosome conformation capture (Hi-C) technology makes the construction of chromosome horizontal assembly possible. However, correction of errors and anchoring of sequences to chromosomes during assembly remain significant challenges.
On September 17, 2024, Wang Yi of Southwest University, as the corresponding author, was inNucleic Acids Research Published online entitled“A deep learning-based method enables the automatic and accurate assembly of chromosome-level genomes”The research paper,The studyA deep learning-based method was developed to AutoHiC solve the challenge of chromosome-level genome assembly by improving continuity and accuracy.
Traditional Hi-C-assisted scaffolding usually requires manual improvement, but AutoHiC the use of Hi-C data for automated workflows and iterative error correction. When trained on data from more than 300 species, the AutoHiC showed robust average error detection accuracy of more than 90%. Benchmark results confirm its significant impact on genome contiguity and error correction.AutoHiC innovative methods and integrated results constitute a breakthrough in automated error detection, providing more accurate genome assembly for advancing genomics research.
Genomics is critical to understanding gene function and evolutionary relationships between species.Genome sequence information at the chromosome level is essential to scientific progress in these areas. This goal is crucial because it provides the basis for uncovering the principles of biological systems and studying the molecular mechanisms of underlying diseases. Recent advances in genome assembly have been driven by the advent of long-read sequencing technologies, such as PacBio's single-molecule real-time sequencing (SMRT) and Oxford Nanopore Technology (ONT). These technological advances overcome the limitations of traditional next generation sequencing (NGS) methods by providing read lengths that exceed traditional boundaries. These factors have led to a significant increase in comprehensive and continuous genome assembly.
However, despite these advances, achieving assembly at the chromosome level remains a challenging goal as current read lengths cannot span the entire chromosome.
High-throughput chromosome conformation capture (Hi-C) sequencing has become an indispensable tool in genome assembly.This technique combines proximity ligation and sequencing, and seeks to assemble contigs scaffolds into chromosome level by exploiting the increased density of Hi-C linkage pairs between adjacent contigs. Many tools have been developed including Lachesis, 3D-DNA, SALSA, YaHS, instaGRAAL, EndHiC, and AllHiC. However, these tools have their limitations and are subject to various influences. Some require a pre-specified number of chromosomes, which presents a challenge to the user. In addition, errors in the output of these tools often need to be corrected manually, thereby prolonging the process and introducing the possibility of human error.This reliance on human intervention hinders the realization of fully automated genome assembly, especially when targeting chromosome-level accuracy of Hi-C data.
In addition to major genome research projects, there is an urgent need for large-scale automation of high-quality chromosome-level genome assembly.However, traditional assembly methods struggle with the size and complexity of growing datasets and face challenges related to algorithmic accuracy and human resource availability. Deep learning has become an integral part of the life sciences, making significant contributions to data analysis and processing. Transformer architecture, an attention-based model designed to process long sequences, has achieved significant success in language processing and has shown versatility in various fields such as image analysis, gene expression, and protein folding. Despite the introduction of software such as DeepC, EagleC, VEHiCLE, DeepLoop, and hicGAN, the potential of Hi-C data in identifying assembly errors remains largely untapped.Big data sets provide opportunities for deep learning to make effective use of Hi-C data.
The study introduces AutoHiC, a scalable and computationally efficient deep learning-based Hi-C assembly error correction method.By utilizing genome-wide chromatin contact data from more than half a million Hi-C images from approximately 300 species, AutoHiC can automatically correct assembly errors and improve the continuity and accuracy of genome assembly. The effectiveness of the AutoHiC recognition and error correction algorithm is proved by the comparative analysis of the pre-and post-adjustment contact heat maps. Our results show that AutoHiC outperforms other software to improve genome continuity. The accuracy of the AutoHiC adjustment was confirmed by comparing the adjusted genome sequence with a telomere-to-telomere (T2T) reference genome sequence, which was highly consistent with the T2T genome sequence. In addition, to determine the broad applicability of the AutoHiC, the study tested on a variety of species, including those with large genomes, numerous chromosomes, and polyploidy.The results confirm that the AutoHiC is applicable and effective in different genome assembly scenarios.
In summary, AutoHiC use deep learning and Hi-C data to automate genome assembly at the chromosome level and improve accuracy. It detects and corrects errors in Hi-C assembly, complements manual management efforts and simplifies genome assembly at the chromosome level.The study hopes that AutoHiC will advance genomics research and deepen the understanding of three-dimensional genome structure and function.
Chongqing Chongfan Technology Co., Ltd. Optical platform film thickness meter OCT optical coherence tomography LabVIEW control
