NucMerge: Genome assembly quality improvement assisted by alternative assemblies and paired-end Illumina reads

Ksenia Khelik, Alexander J. Nederbragt, Geir Kjetil Sandve, Torbjørn RognesbioRxiv 2018 · DOI

Abstract

Despite advancements in second-generation sequencing, genome assemblies often remain fragmented and contain errors. Researchers frequently run multiple assemblers, but selecting a single "best" assembly involves trade-offs. NucMerge is a tool designed to improve assembly accuracy by incorporating information from alternative assemblies and paired-end Illumina reads. It corrects insertions, deletions, substitutions, and inversions while locating rearrangement errors. Benchmarking against Metassembler and GAM-NGS demonstrates that NucMerge consistently improves assembly accuracy across diverse datasets.

Introduction

The proliferation of assembly algorithms like Velvet, SPAdes, and ABySS has provided researchers with various ways to reconstruct genomes. However, these tools often exhibit different strengths; one might offer high contiguity but suffer from frequent substitutions, while another might be highly accurate but fragmented. Traditionally, the "best" assembly is chosen based on metrics like N50, which often necessitates a compromise between accuracy and length.

Assembly reconciliation, or merging, seeks to combine the advantages of multiple assemblies. While several tools exist for this purpose, many suffer from high false-positive rates or fail to address specific structural errors. NucMerge addresses these gaps by focusing specifically on enhancing assembly fidelity through a rigorous multi-stage pipeline.

Methodology

NucMerge operates through a three-stage process: Evaluation, Comparison, and Correction. This workflow leverages both the internal consistency of the assemblies and external evidence from paired-end (PE) Illumina reads.

The NucMerge workflow: Evaluation, Comparison, and Correction.

1. Evaluation Step

The core of NucMerge's reliability lies in its dual-tool evaluation. It utilizes Pilon and NucBreak to assess how well Illumina reads map to both the "target" (the assembly to be improved) and the "query" (the alternative assembly). Pilon identifies discordant read alignments and coverage anomalies, while NucBreak focuses on properly mapped reads and alternative alignments. By combining these tools, NucMerge ensures a comprehensive detection of potential errors.

2. Comparison Step

NucMerge employs NucDiff to perform a base-by-base comparison between the target and query assemblies. This step catalogs every sequence difference, providing the necessary data for targeted corrections.

3. Correction Step

The correction phase identifies regions where the target assembly is predicted to be erroneous but the query assembly is deemed "clean" by the evaluation tools.

  • Local Errors and Inversions: If the query sequence is validated as correct, it replaces the corresponding erroneous region in the target.
  • Structural Rearrangements: For complex inter- or intra-chromosomal rearrangements, NucMerge adopts a conservative strategy. Instead of attempting to resolve the rearrangement—which could introduce new errors—it splits the target sequence at the breakpoint. This prioritizes accuracy over contiguity.

Experimental Evaluation

The authors validated NucMerge using simulated human data and real-world bacterial datasets.

Performance on Real Bacterial Data

In a comprehensive test involving 48 pairs of real bacterial assemblies (including E. coli and S. aureus), NucMerge improved the assembly in 36 cases. In these instances, it reduced existing errors without introducing new ones. In comparison, Metassembler and GAM-NGS improved only 3 and 4 cases, respectively.

Sequential Merging

NucMerge was also tested on the GAGE-B dataset to evaluate its performance when merging more than two assemblies. It was the only tool that consistently avoided "accuracy degradation" during sequential merging, successfully reducing local errors and breakpoints in approximately half of the test cases.

Discussion

A key finding of this research is the power of complementarity. By requiring the query assembly to be validated by both Pilon and NucBreak, NucMerge significantly minimizes the risk of false-positive corrections. Furthermore, the study found that the order of merging matters: adding assemblies in order of decreasing N50 is optimal for local error correction, while increasing N50 order is better for detecting structural errors.

While NucMerge excels at improving accuracy, its conservative approach to rearrangements means it does not currently increase assembly contiguity. Future developments will focus on resolving these rearrangements to merge sequences, thereby improving both the length and the fidelity of the final assembly.

Conclusion

NucMerge provides a robust, accuracy-focused solution for genome assembly reconciliation. By effectively leveraging alternative assemblies and short-read data, it offers a significant improvement over existing state-of-the-art tools, particularly in bacterial genomics.

Summary updated

Abstract

Abstract Background In spite of the major breakthroughs in the second-generation sequencing technologies and the developments of a plethora of assemblers over the last ten years, the resulting genome assemblies may still be fragmented and contain errors. It is typical in genome projects with second-generation reads involved to run multiple assemblers with different parameters and choose the best assembly. However, such an approach is always a trade-off between the strengths and weaknesses of the assemblies. To exploit the advantages of different assemblers, an alternative approach that combines the best parts of several assemblies into one may be applied. The existing tools based on such an approach assist in elongation of assembly fragments and/or improvement of assembly accuracy. Though there has been progress with such a strategy, there is still room for improvement of the existing tools. Results We present NucMerge, a tool for improving genome assembly accuracy by incorporating information derived from an alternative assembly and paired-end Illumina reads from the same genome. The tool corrects insertion, deletion, substitution, and inversion errors and locates different inter- and intra-chromosomal rearrangement errors. NucMerge was compared to two existing alternatives, namely Metassembler and GAM-NGS. Conclusions The benchmarking results show that NucMerge has generally better performance than the other tools tested, providing accuracy improvement of more assemblies. NucMerge is freely available at https://github.com/uio-bmi/NucMerge under the MPL license.
BibTeX
@misc{khelik2018nucmerge,
  author = {Ksenia Khelik and Alexander J. Nederbragt and Geir Kjetil Sandve and Torbjørn Rognes},
  title = {{NucMerge: Genome assembly quality improvement assisted by alternative assemblies and paired-end Illumina reads}},
  year = {2018},
  doi = {10.1101/483701},
  howpublished = {bioRxiv},
}
Figure