Source-linked AI summary
Metagenome assembly of high-fidelity long reads with hifiasm-meta
Xiaowen Feng, Haoyu Cheng, Daniel Portik, Heng Li
TL;DR
Standard metagenome assembly procedures can discard reads or overlaps in ways that create gaps, particularly for low-abundance genomes and alternative haplotypes. Hifiasm-meta addresses these issues through read selection, modified chimera handling, contained-read retention, and coverage-based graph cleaning, with reported improvements in contiguity and gap patching.
Problem
Standard assembly procedures may discard low-abundance or haplotype-specific reads, creating assembly gaps.
Method
Hifiasm-meta combines optional read selection with modified chimera detection, contained-read treatment, and coverage-based graph cleaning.
Results
The modified chimera-detection threshold improves contiguity for low-abundance genomes, while graph cleaning can patch remaining assembly gaps.
Takeaways & Limitations
Retaining reads when alternative haplotypes are plausible prioritizes gap avoidance, while later graph cleaning removes redundant bubble-like structure and resolves selected ambiguities.
Takeaways & Limitations
Read selection assumes the input is not particularly sorted.
Abstract
from arXiv · showhide
Current metagenome assemblers developed for short sequence reads or noisy long readswere not optimized for accurate long reads. Here we describe hifiasm-meta, a new metagenome assembler that exploits the high accuracy of recent data. Evaluated on seven empirical datasets, hifiasm-meta reconstructed tens to hundreds of complete circular bacterial genomes per dataset, consistently outperforming other metagenome assemblers.
Methods
hifiasm-meta combines optional read selection with error correction, overlap handling, graph construction, and graph cleaning tailored to metagenome challenges. Its evaluation uses standardized assemblers, binning, and CheckM-based completeness and contamination assessment.
- Workflow: The workflow comprises optional read selection, sequencing error correction, read overlapping, string graph construction, and graph cleaning.Error correction and read overlapping are largely inherited from hifiasm, while the remaining stages were revamped.
- Read selection: Read selection examines k-mer occurrence percentiles and retains reads when at least one low-abundance signal falls below its threshold.Selection processes reads in batches of 2,000 and checks the 3%, 5%, and 10% percentiles against thresholds 10, 50, and 50.
- Read handling: Hifiasm-meta disables a chimera heuristic for reads whose two ends overlap with five or fewer other reads, improving contiguity for low-abundance genomes.The change avoids discarding reads that may contain statistically uncovered regions because their genomes have low abundance.
- Read handling: Contained reads are retained when exact-overlap evidence suggests different haplotypes, and redundant bubble-like subgraphs are later removed.A contained read is dropped only when no other similar haplotypes appear around it.
- Graph cleaning: For unitigs longer than 100kb, graph cleaning drops one ambiguous overlap when rAB > 0.7 and rAC < 0.7.Here rAB is the minimum coverage of unitigs A and B; the rule uses unitig coverage to prune bifurcating overlaps.
- Evaluation: Assemblies were run with 48 CPU threads and compared with HiCanu v2.1.1 and metaFlye v2.8.1, using read selection for the two mock communities.MetaBAT2 was used for initial binning and circular contigs of at least 1Mb were separated into individual MAGs; CheckM measured completeness and contamination.
Data availability
The study provides accession information for HiFi data and makes generated assemblies and reference resources available through specified repositories.
- Data availability: HiFi datasets were obtained from the NCBI Sequence Read Archive using accession numbers listed in Table 1.Generated assemblies are available at the hifiasm-meta FTP repository.
- Data availability: Reference genomes and evaluation databases were supplied through ZymoBIOMICS, ATCC, CheckM, and GTDB-Tk repositories.The passage lists download locations for ZymoBIOMICS references, ATCC genomes, and the CheckM and GTDB-Tk databases.
Extended Data Figures
Extended Data Figure 1 presents the hifiasm-meta assembly graph for the sheepA dataset, excluding short disconnected contigs from the display.
- Extended Data Figure 1: The figure shows the hifiasm-meta assembly graph for the sheepA dataset.Short disconnected contigs are not shown.