In the ever-evolving landscape of genome sequencing, achieving reference-quality, telomere-to-telomere (T2T) phased assemblies has long stood as an aspirational benchmark. These assemblies offer the most complete and accurate genomic maps, capturing the entire length of chromosomes from one telomere to the other. Traditionally, such endeavors have been both technically challenging and financially prohibitive, especially when scaled across diploid and polyploid genomes. The complexity arises chiefly because assembling complete genomes demands integrating multiple types of sequencing reads, each with their own strengths and limitations.
Conventionally, generating T2T assemblies requires the amalgamation of highly accurate long-read technologies, exemplified by PacBio HiFi sequencing or the now-obsolete ONT Duplex reads. These high-fidelity reads provide the precision necessary to correctly resolve genomic sequences. However, to span complex repetitive regions and achieve chromosome-length contiguity, they must be paired with ultra-long Oxford Nanopore Technologies (ONT) Simplex reads. The combination, while effective, drastically escalates costs and the required input of high-quality genomic DNA, thus limiting broad adoption especially in studies involving numerous or large genomes.
Addressing these obstacles, an innovative approach has recently emerged that leverages deep learning to dramatically enhance the utility of ultra-long ONT Simplex reads. Known as HERRO, an acronym for Haplotype-aware ERRor cOrrection, this novel framework harnesses artificial intelligence algorithms to correct sequencing errors while specifically preserving the subtle differences vital for distinguishing haplotypes or repeated genomic regions. HERRO’s design uniquely acknowledges and utilizes informative polymorphic sites, enabling unparalleled error correction fidelity across complex diploid human genomes.
The underlying principle of HERRO lies in its ability to model the intricate sequence variation inherent in diploid genomes. By incorporating a haplotype-aware correction strategy, HERRO maintains true biological differences rather than erroneously smoothing them out, a common pitfall in conventional error correction algorithms. This precision facilitates a read accuracy increase up to 100-fold, setting a new standard for the quality of ONT Simplex reads. Importantly, this leap in accuracy comparably matches—and in some cases rivals—that of more established yet costlier approaches dependent on multiple sequencing platforms.
Integrating HERRO-corrected ONT reads with Verkko, a cutting-edge de novo genome assembler optimized for long reads, researchers have reconstructed up to 32 chromosomes telomere-to-telomere, including both sex chromosomes X and Y. The assemblies generated demonstrate impressive contiguity, consistently achieving NGA50 values exceeding 100 megabases across multiple human genomes. Such milestones signal a profound shift, as they indicate that ultra-long ONT reads, once considered too error-prone for standalone assembly, can now independently underpin near-complete diploid genome assemblies.
Further underscoring its broad applicability, HERRO supports both major ONT Simplex chemistries—R9.4.1 and the newer R10.4.1—ensuring immediate relevance across current sequencing platforms. Additionally, the method generalizes effectively to non-human species, suggesting avenues for revolutionizing genome assembly in a diversity of biological contexts. This cross-species adaptability opens up possibilities for scaling high-quality genomic studies without prohibitive increases in input material or sequencing complexity.
The implications of HERRO’s development extend well beyond cost reduction. By streamlining the sequencing workflow into a single-platform solution reliant on error-corrected ultra-long ONT reads, the process circumvents the logistical and technical hurdles associated with integrating multiple sequencing technologies. This paradigm not only reduces financial barriers but also minimizes necessary sample handling, which is crucial for studies involving rare or precious biological specimens.
On a technical front, HERRO’s architecture employs deep neural network models trained on extensive genomic datasets, enabling nuanced error recognition and correction while preserving true polymorphisms. The innovation lies in balancing error suppression with biological authenticity, a feat achieved by leveraging sequence context and haplotype structure information during correction. This contrasts with earlier error correction strategies that often indiscriminately corrected mismatches, inadvertently erasing meaningful haplotype differences.
The deep learning framework also adapts dynamically to sequencing chemistry properties, accommodating systematic errors characteristic of different ONT flow cell versions. This adaptability enhances robustness and positions HERRO as a future-proof solution able to keep pace with ONT technology’s rapid evolution.
In practice, implementation of HERRO followed by Verkko assembly demonstrated substantial improvements in contiguity and accuracy benchmarks when applied to diverse human samples. The reconstructions unveiled nearly complete chromosomal assemblies with telomere-to-telomere continuity, setting a new standard for diploid genome resolution. Moreover, the approach effectively uncovered and preserved structural variations and repeat expansions, which are often challenging to resolve but critical for comprehensive genomic analyses.
HERRO thus exemplifies how computational innovation, applied judiciously, can unlock latent potential in existing sequencing chemistries. By transforming traditionally high-error ultra-long ONT reads into highly reliable substrates for assembly, the technology paves the way for expanding high-fidelity, full-chromosome assemblies in both research and clinical domains.
Looking forward, the implications of this technique are vast. Equipping labs worldwide with cost-effective means to produce reference-grade genome assemblies democratizes access to genomic information, accelerating discovery in fields ranging from evolutionary biology to personalized medicine. Further, as sequencing continues its rapid trajectory of advancement, tools like HERRO ensure that data quality improvements keep pace, helping to realize the full promise of genomics.
In sum, HERRO’s haplotype-aware error correction combined with ultra-long ONT read assembly heralds a new era in genome sequencing where completeness and accuracy no longer require prohibitive budgets or complex multi-platform strategies. By harnessing deep learning and the unique strengths of nanopore sequencing, this innovation charts a promising path toward comprehensive, accessible, and affordable chromosome-scale genomics.
Subject of Research: Telomere-to-telomere genome assembly and error correction of Oxford Nanopore ultra-long Simplex reads using deep learning.
Article Title: Telomere-to-Telomere Assembly Using HERRO-Corrected Simplex Nanopore Reads.
Article References:
Stanojević, D., Lin, D., Nurk, S. et al. Telomere-to-Telomere Assembly Using HERRO-Corrected Simplex Nanopore Reads. Nature (2026). https://doi.org/10.1038/s41586-026-10563-y
Image Credits: AI Generated
