Efficient assembly of nanopore reads via highly accurate and intact error correction

Ying Chen, Fan Nie, Shuang Xie, Yingfeng Zheng, Qi Dai, T. A. Bray, Yao-Xin Wang, Jian-Feng Xing, Zhijian Huang, Depeng Wang, Lijuan He, Feng Luo, Jianxin Wang, Yizhi Liu, Chuan‐Le XiaoNature Communications 2021 · DOI

Abstract

Long nanopore reads are advantageous in de novo genome assembly. However, nanopore reads usually have broad error distribution and high-error-rate subsequences. Existing error correction tools cannot correct nanopore reads efficiently and effectively. Most methods trim high-error-rate subsequences during error correction, which reduces both the length of the reads and contiguity of the final assembly. Here, we develop an error correction, and de novo assembly tool designed to overcome complex errors in nanopore reads. We propose an adaptive read selection and two-step progressive method to quickly correct nanopore reads to high accuracy. We introduce a two-stage assembler to utilize the full length of nanopore reads. Our tool achieves superior performance in both error correction and de novo assembling nanopore reads. It requires only 8122 hours to assemble a 35X coverage human genome and achieves a 2.47-fold improvement in NG50. Furthermore, our assembly of the human WERI cell line shows an NG50 of 22 Mbp. The high-quality assembly of nanopore reads can significantly reduce false positives in structure variation detection.

Introduction: The Challenge of Nanopore Errors

Oxford Nanopore Technologies (ONT) provides "ultra-long" reads that are critical for resolving complex, repetitive genomic regions. However, these reads are characterized by high error rates (typically 12–20%) and the presence of "High-Error-Rate Subsequences" (HERS)—localized stretches where error rates significantly exceed the read average.

Characterization of errors in nanopore reads across species. (A) Error rate distribution. (B) Schematic of a High Error Rate Subsequence (HERS). (C) Fraction of reads containing HERS as a function of read length.

Traditional error-correction pipelines often handle HERS by trimming them. While this increases local accuracy, it fragments long reads into shorter pieces, undermining the primary advantage of long-read sequencing and resulting in fragmented genome assemblies. Additionally, the computational cost of standard correction is immense, often requiring tens of thousands of CPU hours for human-scale genomes.

The NECAT Pipeline

To address these limitations, we developed NECAT, a computational pipeline featuring three core innovations designed to preserve read length while achieving high accuracy.

Adaptive Read Selection

Rather than applying a rigid global error threshold for selecting "supporting reads" during correction, NECAT utilizes a dual-threshold approach. By calculating individual thresholds based on each read's specific quality, NECAT ensures that even the most error-prone reads receive sufficient supporting data for correction without introducing noise from false alignments.

Two-Step Progressive Correction

This novel method avoids the "trimming problem" by correcting reads in stages. First, the algorithm identifies and corrects "easy" segments—Low-Error-Rate Subsequences (LERS). In the second step, these newly accurate segments serve as anchors to help resolve the more complex HERS regions. This progressive refinement keeps the original read intact, preserving its scaffolding power.

Two-Stage Assembler and Efficiency

NECAT employs a two-stage assembler that uses initial high-quality contigs as a foundation, then leverages raw nanopore reads to bridge any remaining gaps. To optimize performance, we implemented a Coverage Count Array (CCA) that halts computationally expensive alignments once a region has reached sufficient support (e.g., 12x), drastically reducing redundant processing.

Results and Performance

NECAT was evaluated across a diverse range of species, from E. coli and yeast to complex plants like O. sativa and human cell lines (NA12878 and the WERI retinoblastoma line).

  • Computational Efficiency: NECAT demonstrated a 6.5 to 258-fold speed increase over Canu. A 35X human genome was assembled in just 8,122 CPU hours.
  • Assembly Contiguity: On the human NA12878 dataset, NECAT achieved a 2.47-fold improvement in NG50 compared to Canu, directly resulting from its ability to maintain intact reads.
  • High-Quality Human Assembly: The assembly of the WERI cell line reached an NG50 of 22–28 Mbp, representing a significant benchmark for standard nanopore libraries.
  • Structural Variation (SV) Detection: By producing more contiguous assemblies, NECAT significantly reduces false positives in SV detection, as large-scale genomic changes are less likely to be misidentified as assembly gaps.

Conclusion

NECAT represents a significant advancement in nanopore bioinformatics by prioritizing read "intactness" over simple trimming. By successfully correcting high-error subsequences rather than discarding them, NECAT enables faster, more contiguous, and more accurate genome assemblies, particularly for large and complex genomes like that of humans.

Summary updated

Abstract

Long nanopore reads are advantageous in de novo genome assembly. However, nanopore reads usually have broad error distribution and high-error-rate subsequences. Existing error correction tools cannot correct nanopore reads efficiently and effectively. Most methods trim high-error-rate subsequences during error correction, which reduces both the length of the reads and contiguity of the final assembly. Here, we develop an error correction, and de novo assembly tool designed to overcome complex errors in nanopore reads. We propose an adaptive read selection and two-step progressive method to quickly correct nanopore reads to high accuracy. We introduce a two-stage assembler to utilize the full length of nanopore reads. Our tool achieves superior performance in both error correction and de novo assembling nanopore reads. It requires only 8122 hours to assemble a 35X coverage human genome and achieves a 2.47-fold improvement in NG50. Furthermore, our assembly of the human WERI cell line shows an NG50 of 22 Mbp. The high-quality assembly of nanopore reads can significantly reduce false positives in structure variation detection.
BibTeX
@inproceedings{chen2021efficient,
  author = {Ying Chen and Fan Nie and Shuang Xie and Yingfeng Zheng and Qi Dai and T. A. Bray and Yao-Xin Wang and Jian-Feng Xing and Zhijian Huang and Depeng Wang and Lijuan He and Feng Luo and Jianxin Wang and Yizhi Liu and Chuan‐Le Xiao},
  title = {{Efficient assembly of nanopore reads via highly accurate and intact error correction}},
  booktitle = {Nature Communications},
  year = {2021},
  doi = {10.1038/s41467-020-20236-7},
}
Figure