Head-to-Head Benchmarking of CycloneSEQ versus ONT on the Chinese Quartet Family 1 Reference Materials

DNA double helix passing through a futuristic sequencing microchip. It uses a subtle split-lighting effect (cool cyan/blue on the left, vibrant magenta/purple on the right)

Key Takeaways

Summary

  • •
    Superior Read Length & Consistent Quality: CycloneSEQ S1 delivers longer Normal-long reads (35 kb vs. 24 kb) and more stable base quality across library types (Q21/Q20) compared to the ONT R10 platform.
  • •
    Comparable T2T Assembly Performance: Both platforms achieve near telomere-to-telomere (T2T) assembly quality, generating highly contiguous assemblies with contig N50 > 120 Mb and QV > 49 in hybrid workflows.
  • •
    Reliable Structural Variant Detection: Both platforms enable highly accurate SV calling with F1-scores around 90% (alignment-based) and >92% (assembly-based), effectively supporting complex and rare-disease research.
  • •
    High-Accuracy Methylation Profiling: CycloneSEQ S1 excels in direct DNA methylation detection, achieving a 94% per-read F1-score and a 96% correlation with WGBS gold-standard data.

Comparative Performance Evaluation

A head-to-head benchmark was conducted to systematically evaluate the performance of MGI’s newly released CycloneSEQ S1 chemistry against Oxford Nanopore Technologies (ONT), using the Chinese Quartet Family 1 Reference Materials [1] as the reference standard.

The evaluation covered four key dimensions: sequencing quality control, de novo genome assembly, structural variant detection, and DNA methylation analysis. Detailed results for each dimension are presented in the following sections. Overall, the CycloneSEQ S1 chemistry demonstrates performance comparable to the ONT platform R10 Flow Cell across all evaluated metrics. Platform-specific strengths and trade-offs are discussed in detail below.

Section 1: Sequencing Quality Control

Read length and base accuracy are the two cornerstone metrics of long-read sequencing, jointly determining the quality of downstream assembly and variant calling. Read length enables the spanning of repetitive regions and resolution of structural variants, while base quality defines the reliability of each nucleotide call. Together, these metrics constitute the foundation for all downstream applications of long-read sequencing platforms.

Sequencing samples: Chinese Quartet Family 1 Reference Materials — Daughter LCL5 (D5), Daughter LCL6 (D6), Father LCL7 (F7), Mother LCL8 (M8) Sequencing platforms: CycloneSEQ G400-ER with S1 chemistry, ONT with R10 flow cell and chemistry Library types: Normal-long (NL), Ultra-long (UL)

Quality Control Conclusions
In terms of sequencing read length: Under the Normal-long library type, CycloneSEQ S1 demonstrates an N50 exceeding 35 kb, compared to 24 kb for ONT. Under the Ultra-long library type, the two platforms deliver comparable performance (CycloneSEQ S1: 85 kb; ONT: 91 kb).

Regarding base quality: CycloneSEQ S1 achieves stable Q20+ performance for both library types (Q21 / Q20). For ONT, NL libraries reach Q28 with the SUP basecaller, whereas UL libraries yield Q18 using the HAC basecaller.

Overall, CycloneSEQ S1 exhibits smaller base-quality fluctuations across library types and more consistent quality for ultra-long libraries. Both platforms meet the requirements for downstream

Table 1. Comparison of sequencing quality-control metrics

Platform Library type Read-length N50 (bp) Sequencing quality (modal Q)
CycloneSEQ S1 Normal-long 35k Q21
CycloneSEQ S1 Ultra-long 85k Q20
ONT Normal-long 24k Q28*
ONT Ultra-long 91k Q18#

* ONT data generated using the SUP basecaller.

# ONT data generated using the HAC basecaller.

Section 2: Genome Assembly

A key advantage of long-read sequencing lies in its capacity for de novo genome assembly. Longer reads span repetitive regions more effectively, yielding more contiguous assemblies that better reflect true genomic structures. To systematically compare the CycloneSEQ S1 and ONT platforms, we designed three assembly strategies: nanopore-read-only assembly, multi-long-read hybrid assembly, and combined long-read and short-read assembly.

Strategy 1: Nanopore-read-only Assembly

With ongoing improvements in nanopore sequencing quality and assembly algorithms, nanopore-only assembly offers a cost-effective solution requiring only a single sequencing technology.

  • Sequencing data: 50× CycloneSEQ S1-NL vs. 50× ONT-NL
  • Analysis workflow: Hifiasm [2,3,4]


Assembly Conclusions
Both CycloneSEQ S1-NL and ONT-NL produce highly contiguous assemblies with contig N50 exceeding 80 Mb and QV (quality value) above 35, indicating comparable performance between the two platforms.

Figure 2. Nanopore-read-only assembly comparison: CycloneSEQ S1-NL vs ONT-NL
Figure 2. Nanopore-read-only assembly comparison: CycloneSEQ S1-NL vs ONT-NL

Strategy 2: Multi-long-read Hybrid Assembly

Hybrid assembly combining PacBio HiFi and nanopore ultra-long reads is a mainstream strategy for generating high-quality telomere-to-telomere (T2T) genomes, leveraging the complementary strengths of both technologies.

  • Sequencing data: 50× PacBio HiFi + 50× CycloneSEQ S1-UL vs. 50× PacBio HiFi + 50× ONT-UL
  • Analysis workflow: Hifiasm [2,3,4]


Assembly Conclusions
Hybrid assembly using PacBio HiFi plus CycloneSEQ S1-UL yields contiguity and accuracy comparable to PacBio HiFi plus ONT-UL. Both workflows generate high-quality assemblies with contig N50 > 120 Mb and QV > 49.

Figure 3-1&Figure 3-2. Comparable contiguity and accuracy between PacBio HiFi + CycloneSEQ S1-UL and PacBio HiFi + ONT-UL hybrid assemblies
Figure 3-1&Figure 3-2. Comparable contiguity and accuracy between PacBio HiFi + CycloneSEQ S1-UL and PacBio HiFi + ONT-UL hybrid assemblies
Figure 3. Comparable contiguity and accuracy between PacBio HiFi + CycloneSEQ S1-UL and PacBio HiFi + ONT-UL hybrid assemblies

Strategy 3: Combining Long-read and Short-read Data

Short-read data, with higher single-base accuracy, can further improve assembly quality. Individual short-read data enables post-assembly polishing and substantially improves assembly QV, while pedigree-based short-read data effectively reduces assembly Hamming error.

  • Sequencing data: CycloneSEQ S1-NL vs. CycloneSEQ S1-NL + DNBSEQ (individual short reads / pedigree short reads)
  • Analysis workflow: CycAsm [5] (MGI in-house tool)


Assembly Conclusions

  1. Incorporating individual short-read data for post-assembly polishing substantially improves assembly QV, from QV38 to QV62.
  2. Incorporating pedigree short-read data effectively reduces assembly Hamming error, from 0.39 to 0.03. Across nanopore-read-only assembly, multi-long-read hybrid assembly and combined long-read and short-read assembly evaluations, CycloneSEQ S1 achieves performance comparable to the ONT platform.


Across nanopore-read-only assembly, multi-long-read hybrid assembly, and combined long-read and short-read assembly evaluations, CycloneSEQ S1 achieves performance comparable to the ONT platform.

Figure 4. Pedigree short-read data reduces Hamming error; progeny short-read data markedly improves QV values.
Figure 4. Pedigree short-read data reduces Hamming error; progeny short-read data markedly improves QV values.
Figure 4. Pedigree short-read data reduces Hamming error; progeny short-read data markedly improves QV values.
Figure 4. Pedigree short-read data reduces Hamming error; progeny short-read data markedly improves QV values.

Section 3: Structural Variation Detection

Structural variations (SVs) remain challenging to detect with short-read sequencing, yet constitute critical variants for rare-disease diagnosis and complex-disease research—making them a key strength of long-read sequencing. SVs include insertions, deletions, inversions, and repeat expansions larger than 50 bp, and play important roles in human complex-disease studies, clinical genetic diagnosis, and causal-variant interpretation for rare disorders [6]. Since SVs often involve large-scale genomic rearrangements, long reads can span full variant intervals and provide direct evidence for accurate identification of complex SVs [7].

Using a unified SV-calling pipeline, we evaluated CycloneSEQ S1 and ONT via two strategies: alignment-based SV calling and assembly-based SV calling.

Strategy 1: Alignment-based SV calling

  • Sequencing data: 50× CycloneSEQ S1-NL/UL vs. 50× ONT-NL/UL
  • SV analysis workflow: Minimap2 [8] + Sniffles2 [9]
  • SV truth set: Chinese Quartet CQ1 [1]
 
SV Conclusions
Alignment-based SV calling shows overall comparable performance between CycloneSEQ S1 and ONT. CycloneSEQ S1 delivers slightly higher precision, while ONT achieves marginally higher recall. The overall F1-scores remain at nearly identical levels (~90%), with CycloneSEQ at 90.6% and ONT at 89.8%. 
Figure 5. Alignment-based SV-calling performance comparison between CycloneSEQ S1 and ONT
Figure 5. Alignment-based SV-calling performance comparison between CycloneSEQ S1 and ONT

Strategy 2: Assembly-based SV calling

  • Sequencing data: 50× CycloneSEQ S1-NL/UL vs. 50× ONT-NL/UL
  • SV analysis workflow: SVIM-asm [10]
  • SV truth set: Chinese Quartet CQ1 [1]
 
SV Conclusions
Under assembly-based SV calling, CycloneSEQ S1 demonstrates performance comparable to ONT, with both platforms achieving SV F1-scores above 92%. Overall, both CycloneSEQ S1 and ONT enable reliable breakpoint detection and complex SV discovery, supporting long-read applications including pedigree variant interpretation and complex-disease-associated SV detection.
Figure 6. Assembly-based SV-calling performance comparison between CycloneSEQ S1 and ONT
Figure 6. Assembly-based SV-calling performance comparison between CycloneSEQ S1 and ONT

Section 4: DNA Methylation Detection

Genomic DNA methylation is a key epigenetic mark that regulates gene expression. Traditional bisulfite-conversion-based methylation assays cause sample damage and lose haplotype-phase information. Nanopore long-read sequencing, by contrast, directly detects native 5mC signals from DNA molecules, preserving both sample integrity and haplotype phase—offering a unique advantage for epigenomic research. Using Chinese Quartet Family 1 reference samples (D5, F7, and M8), we systematically evaluated methylation-calling performance at both the per-read and per-site levels.
 
  1. Per-read: Evaluates accuracy of CpG methylation state on individual reads; assessed for high- and low-methylation sites only.
  2. Per-site: Evaluates accuracy of methylation fractions across all CpG sites genome-wide.
 
Methylation Conclusions
  1. At the per-read level, CycloneSEQ S1 demonstrates higher precision, recall, F1-score, accuracy, and ROC-AUC than ONT (Figure 7-2), with F1-scores of 94% vs. 89%.
  2. At the per-site level, methylation fractions from both platforms show high correlation with WGBS (whole-genome bisulfite sequencing, the gold-standard methylation assay) (Figure 7-3). CycloneSEQ S1 yields a higher Pearson correlation coefficient against WGBS (PCC: 96% vs. 88%).

Key Findings

Overall, the CycloneSEQ S1 chemistry achieves performance broadly comparable to that of ONT across all evaluated dimensions. It demonstrates longer Normal-long read length (35 kb vs. 24 kb) and more consistent base quality across library types (Q21/Q20 vs. Q28/Q18). The two platforms deliver similar performance for near-T2T genome assembly and SV detection. CycloneSEQ S1 produces methylation calls with higher accuracy and greater consistency relative to WGBS gold-standard data.

CycloneSEQ S1 achieves competitive results for long-read applications, including sequencing quality control, de novo assembly, SV detection, and direct methylation profiling. As such, it represents a viable technical alternative for applications in complex plant and animal genomics, as well as clinical genetic research.

Data Source

References

  1. Ren L, Duan X, Dong L, et al. Quartet DNA reference materials and datasets for comprehensively evaluating germline variant calling performance. Genome Biol. 2023;24(1):270.
  2. Cheng H, Concepcion GT, Feng X, Zhang H, Li H. Haplotype-resolved de novo assembly using phased assembly graphs with hifiasm. Nat Methods. 2021;18:170-175.
  3. Cheng H, Jarvis ED, Fedrigo O, Koepfli KP, Urban L, Gemmell NJ, Li H. Haplotype-resolved assembly of diploid genomes without parental data. Nat Biotechnol. 2022;40:1332-1335.
  4. Cheng H, Asri M, Lucas J, Koren S, Li H. Scalable telomere-to-telomere assembly for diploid and polyploid genomes with double graph. Nat Methods. 2024;21:967-970.
  5. CycAsm: A hybrid genome assembly pipeline combining CycloneSEQ long reads and DNBSEQ short reads [Computer software]. (2026). GitHub.
  6. Koeppel J, Weller J, Vanderstichele T, Parts L. Engineering structural variants to interrogate genome function. Nat Genet. 2024;56(12):2623-2635.
  7. Del Gobbo GF, Boycott KM. The additional diagnostic yield of long-read sequencing in undiagnosed rare diseases. Genome Res. 2025;35(4):559-571.
  8. Li H. Minimap2: pairwise alignment for nucleotide sequences. Bioinformatics. 2018;34(18):3094-3100.
  9. Paulin LF, Romanek H, Jaryani F, Sedlazeck FJ. Structural variant calling using Sniffles2. Nat Protoc. 2026.
  10. Heller D, Vingron M. SVIM-asm: structural variant detection from haploid and diploid genome assemblies. Bioinformatics. 2021;36(22-23):5519-5521.

Discover More

Sequencers

Explore our cutting-edge sequencers designed for high performance and accuracy.

CECs

Visit our Customer Experience Centres to see our workflows in action and discuss your projects.

Customer Stories

Discover how our customers are achieving remarkable results with our technologies.