henhaobio
Sign in

Genome biologyBioinformaticsPeer-reviewed journal article

Oatk: a de novo assembly tool for complex plant organelle genomes

Chenxi ZhouiD, Max R. BrowniD, Mark L. BlaxteriD, Shane McCarthyiD, Richard Michael DurbiniD

PDFDOIPreprintCodeData
In-depth full-text digest available · Jump to digest ↓

Abstract

Plant organelle genomes, particularly the large mitochondrial genomes with intricate repetitive structures, present significant challenges for assembly. The advent of long-read sequencing technologies provides a transformative opportunity to generate complete genomes, but problems of resolving alternative structures remain. Here we introduce a novel tool for plant organelle genome assembly from high-accuracy long reads. Our method employs a k-mer based assembler for rapid assembly graph construction, integrates a profile HMM gene database for robust organelle sequence annotation, and leverages a new search method to find the best supported path through the assembly graph. We describe high-quality organelle assemblies for 195 plant species and demonstrate improvements over other methods. The assembled genomes provide multiple insights into structural complexity, heteroplasmy, and DNA exchange between organelles.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (28 Oct 2024) · CC BY 4.0

Oatk combines a sparse syncmer assembler, profile HMM gene databases and graph path search to assemble plastomes and mitogenomes from HiFi reads for 195 plant species.

Plant mitogenomes are large, repeat-rich and divergent. Existing organelle assemblers handle them poorly and underuse PacBio HiFi data because they rely on computationally heavy nuclear assemblers, seed-and-extend classification and mostly short reads; Oatk instead uses a fast sparse k-mer assembler, profile hidden Markov model (HMM) classification in place of seed sequences, and copy-number-aware graph resolution.

Key results

  1. ShownOatk assembled both the plastome and the mitogenome for every one of 195 land plants. Nearly all plastomes came out as single circles, and the mitogenomes were a mix of single-circular, multi-circular and non-circular assemblies.

    Both organelles were assembled for all 195 species. 192 of 195 plastomes were single-circular; the three exceptions are the only Hibiscus species in the set, and all break at the same locus. Mitogenomes: 132 single-circular, 29 multi-circular, 34 non-circular. · Assembling organelle genomes for 195 plant…; The mitogenome structures

  2. ShownThe syncasm assembler built graphs as good as MBG's or better, and ran far faster than PMAT, which often failed or misclassified organelle sequences.

    syncasm and MBG built graphs for all species, and Oatk and MBG-pathfinder gave similar results. GetOrganelle did consistently better on syncasm graphs than on MBG graphs. syncasm was slightly faster than MBG on average but used more memory. PMAT failed on 32 species (errors or the 24-hour wall limit) and misclassified sequences in 147 of 163 completed species. · Assembling organelle genomes for 195 plant…; Assembling plant organelle genomes

  3. SuggestedFor graph resolution, hmmannot plus pathfinder beat GetOrganelle for both organelles and on graphs from all three assemblers. The gap was largest for complex mitogenomes, where seed-based GetOrganelle often mislabelled sequences.

    GetOrganelle labelled plastid sequence as mitochondrial (e.g., moss plastome inverted repeats) and mitochondrial sequence as non-organellar. pathfinder handled every graph GetOrganelle failed on. In most species GetOrganelle finished within seconds versus up to a minute for hmmannot plus pathfinder; pathfinder alone typically took a second or less. Table 1 counts are not in the text. · Assembling organelle genomes for 195 plant…

  4. ShownMapping reads onto the assembly graphs showed widespread structural heteroplasmy: both orientations of the plastome single-copy regions coexist in roughly equal amounts, and repeat-mediated alternative configurations are common in mitogenomes, many with nearly balanced abundances.

    Plastomes: in 23 species with inverted repeats up to 20 kb (18 bryophytes, five angiosperms), read counts for the two forms showed no significant departure from a binomial with probability 0.5. Mitogenomes: 299 vs 301 reads (Pseudognaphalium luteoalbum), 328 vs 361 (Calluna vulgaris direct-repeat forms). Of 165 repeats across 109 genomes, nearly all were heteroplasmic. · The plastome structures; The mitogenome structures; Mapping reads to assembly graphs

  5. ShownDNA transfer between the two organelles is common and mostly runs from plastid to mitochondrion, but none was found in the bryophytes. It appears to be ongoing, and transferred mitochondrial psaA copies have diverged further than their plastid paralogs.

    576 shared segments (at least 1 kb, at least 90% identity) totalling 1.57 Mb in 144 species, ranging from 1 kb to 17.3 kb. Of the 1.10 Mb annotated as genes, 1.09 Mb (98.7%) are plastid genes. Identities span 90% to 100%. Mitochondrial psaA copies lie farther from the most recent common ancestor in 50 of 64 pairs (Wilcoxon signed-rank p = 8.969e-06). · Sequence transfer between organelle genomes; Identifying shared sequences between plastid and…; Constructing the phylogenetic tree for the psaA…

Methods, data and reproducibility

syncasm builds a sparse de Bruijn graph from homopolymer-compressed closed syncmers (k = 1001, s = 31) in HiFi reads, corrects error syncmers and drops low-coverage nuclear unitigs. hmmannot classifies components by profile HMM gene hits, and pathfinder estimates copy numbers by expectation-maximization and searches the graph for a primary circular or linear path.

  • PacBio HiFi whole-genome data for land plants from the Sanger Tree of Life programme, mostly Darwin Tree of Life: 24 monocots, 154 eudicots, 16 mosses, one liverwort · land plants (multiple species) · n = 195 species
  • Oatk plastid and mitochondrial genome assemblies for all species, deposited on Zenodo · land plants (multiple species) · n = 195 species · 10.5281/zenodo.13952353
  • NCBI nucleotide GenBank organelle sequences (Embryophyta, txid3193) used to build plant plastid and mitochondrial profile HMM gene databases · land plants (Embryophyta) · n = 130 plastid and 81 mitochondrial gene profiles
  • HMM gene databases for animal clades (Mammalia, Actinopterygii, Aves, Lepidosauria, Insecta), used to assemble animal mitogenomes as a demonstration · animals (mammal, fish, bird, lizard, insect) · n = five animal species
  • Gene order of the Arabidopsis thaliana reference plastome, used to choose the orientation of plastome single-copy regions · Arabidopsis thaliana · n = 71 protein-coding genes
  • psaA paralog pairs from plastid-mitochondrial shared sequences, used for the phylogenetic divergence analysis · land plants (multiple species) · n = 64 paralog pairs (128 sequences)
Methods & tools
Oatk (syncasm, hmmannot, pathfinder), OatkDB profile HMM gene databases, MBG 1.0.16, PMAT 1.5.3, GetOrganelle 1.7.7.1, HMMER 3.4 (hmmbuild, hmmpress, nhmmscan), MAFFT 7.505, GenomeScope 2.0, GraphAligner 1.0.19, NUCmer and delta-filter 4.0.0rc1, IQ-TREE 2.3.4 with Biopython 1.83 Bio.Phylo, Entrez Direct (esearch, efetch)
Software & environment
MBG 1.0.16, PMAT 1.5.3, GetOrganelle 1.7.7.1, GenomeScope 2.0, HMMER 3.4, MAFFT 7.505, NUCmer/delta-filter 4.0.0rc1, GraphAligner 1.0.19, IQ-TREE 2.3.4, Biopython 1.83. Oatk used the Embryophyta databases (-p, -m), -c at five times haploid coverage and -t 6, defaults otherwise.
Assay conditions & compute
Oatk, MBG, PMAT and GetOrganelle were each given six threads. PMAT was capped at a 24-hour wall time and run on data downsampled to 3× coverage. In most species GetOrganelle finished within seconds; hmmannot plus pathfinder took up to a minute. Hardware and memory figures are not given.
Reproducibility gaps
Raw read accessions are listed only in Supplementary Table S3. Benchmark values (Table 1) and per-species runtime and memory (Supplementary Table S4) are not in the main text, and no container or Oatk version is given.
Limitations (6)
  • Some of the 34 non-circular mitogenomes probably reflect incomplete assembly rather than true linear molecules. Only some, such as Alnus glutinosa, were confirmed by read mapping. · stated by the authors
  • A primary assembly alone omits coexisting conformations unless the assembly graph is also consulted; listing every conformation instead brings redundancy and combinatorial complexity in complex mitogenomes, and linear INSDC records overlook heteroplasmy. · stated by the authors
  • All three Hibiscus plastomes break at the same locus, and the authors cannot tell whether this is a genuine linear structure or a sequencing artifact. · stated by the authors
  • HiFi read length restricted the plastome orientation test to species with inverted repeats up to 20 kb; for mitogenomes, only repeats of at most 10 kb with flanks of at least 10 kb were analysed, to ensure read mapping quality. · stated by the authors
  • Claims that Oatk outperforms other tools are not checked against an independent truth set, such as curated or orthogonally validated organelle genomes, and the Table 1 metrics are absent from the text, so the size of the accuracy gains is hard to judge. · noted in this summary
  • PMAT ran on data downsampled to 3× coverage, its classifications were discarded and its two graphs were merged. Some of its failures may therefore reflect the benchmarking setup rather than the tool used as intended. · noted in this summary

Read this if you assemble plant plastid or mitochondrial genomes from PacBio HiFi data, or study organelle structural heteroplasmy and plastid-to-mitochondrion DNA transfer across many species.

Citations
237
FWCI
70.84
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISJin '20Xian '24Syme '21Zhou '22Freudenthal '20Song '24
76% semantic matchBioinformatics
Jian‐Jun Jin, Wen-Bin Yu, Junbo Yang+4 more·Genome biology·Sep 10, 2020· 4,187 citations