Metagenome assembly of high-fidelity long reads with hifiasm-meta

Abstract

De novo assembly of metagenome samples is a common approach to the study of microbial communities. Current metagenome assemblers developed for short sequence reads or noisy long reads were not optimized for accurate long reads. We thus developed hifiasm-meta, a 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.

Extended Data Figure 1.

The hifiasm-meta assembly graph of the sheepA dataset. Short disconnected contigs are not shown.

Extended Data Figure 2.

Yak QV score correlated with contig coverage. Plots shwoing >1Mb contigs in sheepB assemblies. Contig coverage was estimated by jgi_summarize_bam_contig_depths from metabat2, and alignment was done with minimap2 -a -k 19 -w 10 -I 10G -g 5000 -r 2000 –lj-min-ratio 0.5 -A 2 -B 5 -O 5,56 -E 4,1 -z 400,50. Hifiasm-meta assembled 37 contigs without k-mer errors. HiCanu and metaFlye each assembled 4 contigs without k-mer errors.

Figure 1.

Metagenome assemblies of empirical datasets. (a) Quality score of contigs longer than 300kb from the hifiasm-meta assembly of sheepA. The quality score of a MAG is defined as “completeness −5 × contamination” based on CheckM reports. (b) CheckM evaluation. A MAG is “near-complete” if its CheckM completeness is ≥90% and its contamination level ≤5%, is “high-quality” if completeness ≥70% and contamination ≤10%, or is “medium-quality” if its quality score is ≥50%. “HumanPooled” represents the co-assembly of all four human gut samples. (c) Composition of long contigs in the hifiasm-meta co-assembly of four human gut samples. Each bar represents a contig of ≥1Mb in length. It gives the number of reads used in the contig. Each color corresponds to a human gut sample. A cross at the top of a bar indicates the contig being circular. (d) Phylogeny of human gut MAGs from the co-assembly. A colored clade corresponds to a phylum inferred by GTDB-Tk^,. A MAG is omnivore/vegan-specific if 90% of reads in the MAG come from omnivore/vegan samples, respectively. Two adjacent leaves coming in the same genus have the same shade on the “Genus” outer ring.